EPA420-R-93-017
                                      December 1993
Final Regulatory Impact Analysis
    for Reformulated Gasoline
            Office of Mobile Sources
        U.S. Environmental Protection Agency

-------
                                TABLE OF CONTENTS
                                                                               Page

I.      Treatment of Ethanol	  4
       A.     Exclusion from Market	  4
       B.     Congressional Intent 	  9
       C.     Waiver  	   11
       D.     Reactivity  	   18
       E.     Benefits of Ethanol to the Nation 	   25
       F.     Feasibility of Renewable Oxygenate Proposal	   27
       G.     Emission Impacts of the Renewable Oxygenate Proposal 	   29
       H.     Reg Neg	   32
       I.      Appendices	   33

II.      Simple Model	   61
       A.     Simple Model Equations and Referenced Work  	   61
       B.     Technical Corrections 	   69
       C.     Caps on Fuel Parameters  	   70
       D.     Averaging and Trading	   71

III.     Complex Model Baselines  	   73
       A.     Baseline Fuels 	   73
       B.     Baseline Emissions 	   77
       C.     Appendices	108

IV.     Complex Model	113
       A.     Exhaust Emissions Model  	114
       B.     Complex Model for Non-Exhaust Emissions  	162
       C.     Determining Emission Performances	179
       D.     Range and Extrapolation of the Model	187
       E.     Use of the Complex Model Under Non-Summer Conditions	227
       F.     Behavior of Complete Complex Model	232
       G.     Vehicle Testing	249
       H.     Appendices	275

V.      Environmental and Economic Impact of Phase I Reformulated Gasoline  	303
       A.     Introduction	303

VI.     Phase II Reformulated Gasoline Standards and NOx Standards for Reformulated
       Gasoline 	311
       A.     Introduction	311
       B.     Scientific Justification for NOx Control	313
       C.     Derivation of Fuel Parameter Control Costs  	326
       D.     Derivation of Fuel Parameter Interrelationships	348
       E.     Determination of Emissions Reductions   	367
       F.     Cost-Effectiveness Determination 	373
       G.     Comparison with Other RFC Cost-Effectiveness Analyses	382
       H.     Factors to be Considered in Setting Phase II RFC Standards	383
       I.      Standards	393
       J.     Appendix VI-A: Tables VI-A1 to VI-A6	398

VII.    Anti-Dumping Reguirements for Conventional Gasoline-RIA  	412
       A.     Introduction	412
       B.     Emission Reguirements	412

-------
       C.     Individual Baseline Determination  	416
       D.     Baseline Auditor  	458
       E.     Baseline Submission and Approval 	464

VIM.    Compliance with the Regulatory Flexibility Act  	469
       A.     Determination and Justification of Small Entity Impacts  	469
       B.     Summary and Assessment of Issues and Comments	471

IX.     State Opt-in Provisions  	479
       A.     General Provisions	479
       B.     Phase I Opt-in only	479
       C.     Attainment Area Opt-in	480
       D.     Opt-out	481
       E.     Opt-In Application	481
       F.     Opt-In Lead Time	481
       G.     Define Domestic Capacity  	482
       H.     Opt-in only after Stage II controls and enhanced I/M	482

-------
      Treatment of Ethanol

      One of the most controversial issues in the reformulated gasoline
rulemaking has been the treatment of ethanol blends.  In the April, 16, 1992
proposal, EPA treated all fuels equally by requiring them to meet the same
emission performance standards.  The February 26, 1993 proposal, however,
proposed an incentive program for the use of renewable oxygenates such as
ethanol which would allow them to meet a less stringent performance standard.
Numerous comments were received on a number of topics related to whether EPA
should or should not provide special treatment for ethanol blends.  The
following discussion presents EPA's summary and analysis of those comments
received which EPA believed were relevant to the reformulated gasoline
rulemaking.  Some other comments were received which did not have any
relevance to the rule (e.g., relating to ethanol's use as a neat or near-neat
alternative fuel in alternative fueled vehicles), and as a result did not
influence the decision making process.  A large number of comments were also
received which made unsubstantiated statements regarding the reformulated
gasoline proposal.  To the extent these statements had any bearing to the
final rule they are also addressed in the summary and analysis of comments
below.

A.    Exclusion from Market

      A number of comments were received from ethanol and grain producing
industries that the April 16, 1992 proposal effectively excluded ethanol
blends from participating in the reformulated gasoline market.  Ethanol is
typically splash blended into finished gasoline obtained at the gasoline
terminal, increasing the volatility of the resulting blend.  According to the
commenters the simple model's reliance on a straight RVP standard, with no
volatility waiver for ethanol blends, would require ethanol blenders to use
sub-RVP gasoline blendstocks to enable the resulting blend to meet the simple
model RVP standard.  This would be a clear departure from past practice, where
ethanol could lawfully be splash blended into almost any finished gasoline  (as
long as it did not already contain an oxygenate).  Commenters claimed that
refiners would be unwilling to provide sub-RVP clear gasoline blendstocks for
blending with ethanol given the refiners competitive interest in minimizing
ethanol's market share.   In addition, commenters claimed that limitations in
terminal storage capacity would also preclude the use of unique blendstocks
such as a sub-RVP clear gasoline.  Commenters also noted that the proposal's
provisions on presumptive liability and contract requirements between refiners
and blenders would make provision of the sub-RVP blendstock too risky.
Finally,  even though volatility was only an issue during the high ozone
season, commenters claimed ethanol would as a practical matter be excluded
year round as refiners would likely establish long term contracts for
oxygenates that would make it unlikely for them to switch oxygenates during
the course of the year.   In addition, it would be hard for marketers to switch
from an ethanol based ether  (ETBE) for summer use and an alcohol  (ethanol) for
winter use.  Exclusion from the summer market would therefore lead to
exclusion year round and would jeopardize the ethanol industry's ability to
maintain a market even in the wintertime oxygenate program in CO nonattainment
areas.  While the reformulated gasoline standards appeared to be fuel neutral
on their face, commenters believed that in practice they would lead to
exclusion of ethanol from the reformulated gasoline program.

      As discussed below in subsection E. the commenters also claimed that
this exclusion of ethanol would reduce competition in the oxygenate market and
thereby cause higher prices for reformulated gasoline, reduce investment in
rural and farm communities, increase federal grain subsidies, reduce domestic
jobs, and otherwise have serious adverse effects on the ethanol and grain
producing industries.

-------
      Other commenters  disputed  these various assertions.  For example,
various refiners claimed that  there was no foundation for the assertion that
ethanol would be excluded.   Certain refiners noted that they have used ethanol
in the past and intended to  continue to do so in the reformulated gasoline
program.  Some commenters claimed  ethanol use would in fact expand,  based on
the need for oxygenates year round in the reformulated gasoline program and in
wintertime program in many CO  nonattainment areas.  The Department of Energy
commented that the reformulated  gasoline and wintertime CO programs  were
expected to cause a substantial  increase in the demand for all oxygenates,
including ethanol.

      In general the comments  which argued ethanol would be excluded from the
market focussed on the  possible  reduction in ethanol market growth in the
future under the reformulated  gasoline program, not a reduction in the current
size of the ethanol market.  As  shown in Appendix I.A., a minority of the
areas currently covered by the reformulated gasoline program  (including areas
opted in as of November,  1993) had any gasohol market share in 1990.  Of those
that did, only a few had gasohol market shares above the national average of
approximately 8%.  As a result,  only about 20% of the ethanol demand in 1990
was used in reformulated gasoline  areas, and only about 10% was used in those
areas during the summer VOC  control period.  Even if ethanol achieves only a
small market share under the reformulated gasoline program, it is likely to be
greater than the 4.5% average  market share of gasohol in those areas prior to
the reformulated gasoline requirements, and no economic hardship to  the
industry should result.  Furthermore, demand for ethanol has already risen
dramatically as a result of  the  wintertime oxygenated fuels program  required
under section 211(m)  of the  CAA.1  As a result,  it is  very difficult to
believe any case where  ethanol demand would decrease below the level that
existed prior to the CAA amendments of 1990.  That being said, EPA has always
maintained that ethanol blends would not be excluded from the RFG market by
the April 16, 1992 proposal  and  that demand for ethanol would grow
considerably as a result of  the  program.

      The largest part  of the  cost associated with Phase I reformulated
gasoline is the oxygen  content required to be added by the Act.  The RVP
control cost is small in comparison to the oxygen cost.  Since ethanol costs
less than MTBE per gallon and  contains almost twice as much oxygen per gallon,
it has a considerable economic advantage  (as shown in Appendix I.B),
particularly given various state subsidies, and thus should be able  to expand
its market.  This is particularly  true during the 6.5 to 7.5 months  of the
year when no VOC control is  necessary.  But even during the 4.5 to 5.5 summer
months the only significant  economic issue is the incremental costs  for
distribution and segregation of  ethanol blends which are no different from
those in the winter.  Even the additional RVP control costs for ethanol are
reduced by the oxygen averaging  provisions of the reformulated gasoline
program which allow ethanol  to be  blended at 3.5 wt% oxygen instead  of 2.0 wt%
and receive economic compensation  for the additional oxygen.  Thus,  EPA
believes that the economic advantage of ethanol over other oxygenates will
translate into refiners supplying  sub-RVP gasoline for blending with ethanol
out of economic necessity.   If they choose not to, at least in the areas where
the incremental distribution costs are small or where state subsidies for
ethanol exist, they will place themselves at a considerable disadvantage
relative to their competitors.   The economic advantage of ethanol relative to
      1  Memorandum from Paul Argyropoulos,  Acting  Chief,
Regional/State/Local Coordination  Section,  to Paul Machiele,
"Estimates of  Volume of  Ethanol Use in Oxy  Fuel  Program Areas
During  the 1992/1993 Control  Season,"  August 24,  1993.

-------
MTBE was supported by a recent  analysis  for  the National Corn Growers
Association2  even  with  the higher cost of a sub-RVP blendstock.

      The economic advantage of ethanol  was  very  evident during the 1992-3
oxygenated fuels program where  ethanol made  large market gains, even in areas
which had traditionally used little  or no  ethanol, and ethanol producers
earned record profits.   While some ethanol producers may find it necessary to
reduce their profit margin somewhat  to remain  competitive under the
reformulated gasoline program,  this  neither  excludes ethanol from the
marketplace nor treats  it unfairly relative  to any of the other oxygenates.

      Some commenters claimed that regardless  of  ethanol's economic advantage
refiners would somehow band together to  cut  ethanol out of the reformulated
gasoline market.  EPA believes  such  an action  by  the refiners to be highly
unlikely.  Not only have refiners not taken  such  an action in the past under
either the conventional gasoline market  or under  the oxygenated fuels program,
but comments have also  been received from  various refiners stating their
intent to blend ethanol into reformulated  gasoline.  Furthermore, the
competitive economic risk is simply  too  great  if  even one refiner decided to
break ranks and blend ethanol.   If,  however, refiners were to decide as a
group not to blend ethanol,  nothing  short  of a legal mandate for the use of
ethanol in reformulated gasoline could prevent such action.  EPA has no legal
authority under the CAA to provide such  a  mandate for the use of ethanol.

      Regardless of whether refiners intended  to  exclude ethanol from the
reformulated gasoline market or not,  some  commenters believed this would occur
automatically due to the risk to the refiner and  necessary contract
requirements which would result to ensure  that the final gasoline blend met
the requirements for reformulated gasoline.  While EPA believes that there may
be some additional risk to the  refiners  to rely on downstream blenders for the
final quality of their  reformulated  gasoline,  EPA has provided mechanisms
whereby refiners can minimize or even eliminate such additional risk (refer to
§80.69 of the regulations).   As such, EPA  does not believe it to be a
significant enough risk such that they would avoid entering into contractual
agreements with downstream blenders.

      In addition to being economically  competitive as a pure additive,
ethanol may also be economically competitive as an ether feedstock.  Due to
their low blending vapor pressure, ETBE  (ethyl tertiary butyl ether) and TAEE
(tertiary amyl ethyl ether),  ethers  made from  ethanol, may become competitive
with other oxygenates during the summer  months under the reformulated gasoline
program, especially in  the future when the more stringent Phase II performance
standards take effect.   In fact, some commenters  stated that ETBE was already
cost competitive with MTBE.   Contrary to comments received, the current lack
of ETBE or ETAE production is not solely a result of their relative cost.
Such oxygenates and their production on  a  large scale are very recent
technological developments.   As such there has been no reliable cost
information for them until recently.  Furthermore, their cost effectiveness in
reformulated gasoline is expected to exceed  that  in conventional gasoline due
to their low blending vapor pressure, high octane value per unit oxygen, and
various other factors which are of value to  refiners in meeting the
reformulated gasoline requirements,  but  are  not of value otherwise.  As shown
in Appendix I.E.,  when  these factors are taken into account ETBE nears the
point of being cost competitive with MTBE  even at today's ethanol costs.
      2     "Evaluation of the  USDA Alternative Renewable
Oxygenate Proposal," Final  Report, Jack Faucett  Associates for
the  National  Corn  Growers Association,  October  30,  1993.

-------
      Some commenters claimed that if ethanol was unable to maintain a large
market share during the summer months, that it would become uneconomical to
use during the winter months as well.  They believed refiners would be
unwilling to switch off between various oxygenates.  This belief, however,
appears to be unfounded.  First of all, it is based on the invalid assumption
that ethanol will have no market either in the RFG program or in conventional
fuel areas during the summer months.  Secondly, contrary to the intent of
various refiners to blend ethanol as stated in their comments, it assumes no
refiners would enter into long term contracts to blend ethanol, and third,  it
ignores the fact that refiners have been willing to switch off between
oxygenates in the past.  During the 1992-3 oxygenated fuels program refiners
blended ethanol during the winter and then switched back to conventional
gasoline during the summer.  Furthermore,  many refiners switched between
oxygenates during the winter period as well based on the local economic
conditions.  Ethanol and MTBE producers alike were able to produce and store
adequate oxygenate to meet the peak demand during the winter months.  In
addition, contrary to the commenter's belief, it could be argued that it is
easier for ethanol blends to tolerate an interrupted market than MTBE blends.
The nature of ethanol production is that it could be reduced during the summer
by optimizing on plant production of corn sweeteners which are in high demand
during the summer.  As a result, while ethanol production may be reduced in
lieu of reducing its price to maintain market share, the ethanol production
plant may be able to maintain full operating capacity.  Methanol used to
produce MTBE has no other compensating market during the winter and is forced
to either reduce its price to maintain market share, rely on storage, or shut
down production.  Since the winter period when ethanol is at an economic
advantage is longer than the summer period when MTBE may be at an economic
advantage, ethanol should have less of a burden.  In fact, some of the
statements made supporting the claim that ethanol would be excluded from the
RFG market:  inadequate tankage for multi-grades of fuel, inability of
refiners to switch off between oxygenates, etc., are equally applicable if  not
more so to MTBE.  Furthermore, using ethanol to produce ETBE during the summer
when ETBE's other advantages have value could be a viable means of maintaining
ethanol production year-round.

      None of the comments received provided any data, information, or
analyses to support the statements being made.  As a result, EPA stands behind
its analysis and belief that ethanol is fully capable of competing in the
reformulated gasoline market either as gasohol or as an ether feedstock.  This
is particularly the case during the initial years of the reformulated gasoline
program when ethanol will have to play a strong role due to the demand for
oxygenates under the program which cannot be met through the use of MTBE
alone.  This belief was echoed by a variety of other commenters.

A.    Congressional Intent

      A number of comments were received stating that it was the intent of
Congress in the Clean Air Act (CAA) that ethanol be included in the
reformulated gasoline program and that the April 16, 1992 proposal by EPA
violated that intent by discriminating against ethanol.

      Congressional intent in establishing the reformulated gasoline program
can best be learned from the text of the statute and the legislative history.
A review of the terms of section 211(k) makes clear that the central,
overriding purpose of this provision was the achievement of significant
reductions in ozone forming VOCs and toxic air pollutants in the major
metropolitan areas confronted with the most severe air pollution problems.
Section 211(k)(1) establishes this as the central focus of this program,
mandating that EPA require the greatest achievable reductions in these
pollutants, considering various factors such as cost, air quality and other
impacts.  In addition to this broad expression of intent, Congress specified
certain minimum elements of the reformulated gasoline program.  Section

-------
211(k)(1)  established various restrictions on gasoline  qualities such as
oxygen and benzene  content, and a cap on NOx emission increases, and section
211(k)(3)  established minimum standard levels of performance  for VOC and
toxics emission  reductions.  Congress also addressed a  variety of other
important  elements  of the reformulated gasoline program,  from credit programs
to certification and opt-in by states.

      All  of the minimum requirements in section 211(k)  are set in a fuel
neutral  manner,  using performance criteria that are independent of the type of
gasoline subject to the criteria.  There is no indication from the text of the
statute  that Congress intended that ethanol blends not  be subject to the same
minimum  standards applicable to all reformulated gasoline.  This is in marked
contrast to section 211(h), where Congress quite clearly established one RVP
standard for gasolines  in general, and a separate, less stringent RVP standard
for ethanol blends  under specified circumstances.

      The  legislative history of this provision does contain  many discussions
pertaining to ethanol and other oxygenates.  Since many of the detailed
provisions of section 211(k) were first adopted in floor amendments to the
House and  Senate Bills, much of that legislative history is found in the floor
debate in  the House and Senate on their bills and on the conference committee
bill. A large part of  the discussion there on ethanol  shows  the belief of
many congressmen that ethanol would play a major role in supplying the oxygen
content  required for this program, and would help to clean up gasoline by
reducing toxic air  pollution.3  There  were  no indications, however,  that
ethanol  blends of reformulated gasoline were not subject to the minimum
requirements established for all reformulated gasolines.   There are in fact
indications that Congress saw these as fuel neutral standards that all blends
would have to face.4

      In sum,  EPA believes Congressional intent was to  achieve significant
reductions in important air pollutants in areas of the  Country with some of
the worst  air pollution problems, and intended to accomplish  this in part by
establishing minimum standards of performance that were fuel  neutral and that
all reformulated gasolines would have to meet.  EPA believes  its final rule
faithfully implements this intent by setting identical  emission performance
standards  for all reformulated gasolines, regardless of the oxygenate used.

      It is also true that many congressmen expected that ethanol blends would
be able  to meet  these standards, and would thereby play a major role in the
program.  EPA believes  this was a well founded expectation, and as noted
above,  believes  that ethanol blends will in fact play an important role in
this program.  However, there is no indication that Congress  intended or
authorized EPA to dilute the minimum requirements of the section 211(k)
described  above  to  ensure an important role for ethanol.   Congress did
envision that ethanol would play a large role in the reformulated gasoline
program, but only if it could meet the minimum requirements applicable to all
reformulated gasoline.

      As stated  above,  EPA believes that the demand for ethanol will grow as a
result of  the reformulated gasoline program as it has under the oxygenated
      3   See  e.g.,  136  Congressional Record  53510,11  (March  29,
1990)   (Statement by Senator Daschle);  136  Congressional  Record
H2852   (May 23,  1990)   (Statement by  Representative  Richardson).

      4   See  e.g.,  136  Congressional Record  S3513  (March 29,
1990)(Statement by  Senator Daschle);  136  Congressional Record
S16922  (October 27,  1990)  (Statement  by  Senator Durenberger).

-------
fuels program.   While  it  is  true  that the demand for ethanol could increase
even more if special provisions or  incentives are provided for its use such as
proposed in the February  1993 proposal, EPA does not now believe that such
provisions are  consistent with the  minimum requirements that Congress intended
for all reformulated gasoline.  While EPA is not opposed to further expanding
demand for ethanol, the authority granted to EPA under Section 211(k)  of the
CAA is limited  in this respect.   It would be much more appropriate to provide
such incentives for ethanol  outside of the reformulated gasoline rulemaking as
is currently done with the federal  tax incentives for ethanol.

A.    Waiver

      A large number of commenters  stated that EPA should grant a 1.0 psi RVP
waiver to ethanol blended reformulated gasolines.  They argued that it was the
intent of Congress that the  RVP waiver granted to ethanol blended with
conventional gasoline  under  Section 211(h) of the CAA also apply to ethanol
blended in reformulated gasolines under Section 211(k).  They further argued
that the rationale used by Congress and EPA in granting the waiver under
Section 211(h)  was equally applicable to reformulated gasoline and that the
ozone reactivity and other benefits were sufficient to offset any emission
increase which  might result.  Granting a waiver, it was stated, would allow
ethanol to effectively compete in the reformulated gasoline market.   Finally,
some commenters explained that during the regulatory negotiations on this
rulemaking they had assumed  that  a  one psi RVP waiver would be applied to
ethanol blends  of reformulated gasoline.

      There were also  many comments disputing the above assertions.   These
commenters claimed that a one psi waiver for ethanol blends was unlawful and
that granting such a waiver  would violate the intent of Congress.  They
claimed that a  waiver  would  have  serious adverse environmental impacts.  They
argued that such a waiver would be  discriminatory and unfair, and would be
counter to free market principles.  Finally, they argued that such a waiver
was discussed during the  regulatory negotiations, at least informally, and
that granting an RVP waiver  for ethanol blends would violate the consensus
agreement reached during  these negotiations.

1.    Intent of Congress

      EPA has closely  considered  the arguments that the 1.0 psi waiver for
ethanol blends  provided under section 211(h) applies to the RFC program.  For
the reasons set forth  in  a legal  opinion issued by EPA's legal council,5 EPA
believes that the 1.0  psi waiver  under section 211(h)  does not apply to
section 211(k).

2 .    Environmental Impact

      EPA believes that granting  a  1 psi RVP waiver for ethanol blends would
have significant adverse  environmental impacts.  As shown in a memorandum and
      5  Memorandum  from Acting  General Councel  to  the
Administrator,  dated November 17,  1992,  "Applicability of  One-
psi  Ethanol Waiver  Under Section  211(k)  of  the Clean  Air Act to
Reformulated  Gasoline."

-------
subsequent  letter from EPA staff in the docket,6 7 an ethanol blend that meets
all of the  requirements for reformulated gasoline except that it  has an RVP
1.0 psi higher than that required of non-ethanol blends would cause roughly a
20% increase  in VOC emissions relative to the baseline gasoline  (including
estimates  for the effect of commingling and distillation).   This  represents a
31% increase  in VOC emissions compared to a fuel which meets the  minimum
requirements  for reformulated gasoline.  Based on the relative costs of the
various oxygenates as shown in Appendix I.E., ethanol blends with a 1.0 psi
RVP waiver  would have a significant competitive advantage over other
reformulated  gasoline blends and could be expected to dominate the RFG market
in some areas, particularly mid-western RFG cities such as  Chicago and
Milwaukee where ethanol distribution costs are relatively small.   With a 1.0
psi waiver  EPA expects there would be a significant shift to ethanol blends
and a resulting significant increase in VOC emissions over  what would occur
without a waiver.  Many areas would achieve significantly smaller reductions
in VOC emissions than the minimum 15% required under the Act,  and in areas
with a large  market share for ethanol blends, the reformulated gasoline
program would actually increase summertime VOC emissions and not  decrease
them.  For  example, as shown in Table I.I., even at the pre-program ethanol
market share  of approximately 8% an ethanol waiver would eliminate nearly 28%
of the VOC  control otherwise achieved by the program.  If the ethanol market
share increases to 24% as under 1992-3 the winter oxygenated fuels program  (on
average), roughly 74% of the VOC control otherwise achieved by the program is
eliminated.   At an ethanol market share above just 35%, as  would  be expected
in many areas where a waiver granted (particularly the Midwest as occurred
under the oxygenated fuels program),  all of the VOC control of the program is
eliminated  and VOC emission increases would actually occur  relative to the
pre-program baseline.
      6      Memorandum  from Paul A.  Machiele,  Fuel Studies  and
Standards  Branch,  to  Richard D.  Wilson,  Director,  Office  of
Mobile Sources,  "Update of  the Relative  Ozone Reactivity  of
Reformulated Gasoline Blends," June 11,  1993.

      7      Letter  from Chester J.  France,  Director,  RDSD to Dr.
Gary Whitten, Chief Scientist, Systems Applications
International,  September 24,  1993.

                                    10

-------
Table I.I.  Loss in In-Use VOC Emission Control as a Result of Ethanol Waiver
            Under RFC*


% VOC Reduction
Due to MTBE Share
% VOC Reduction
Due to Ethanol Share (with
distillation)
% VOC Reduction
Due to Comminqlinq
Total % VOC Reduction
Percentage Change from 100%
MTBE Baseline
Ethanol Market Share
0%
11
0
0
11
NA
8%
10.1
-1.0
-1.2"
7.9
28
24%
8.4
-3 .1
-2.4
2.9
74
30%
7 .7
-3 .9
-2.4
1.4
87
35%
7.2
-4.6
-2.4
0.2
98
      Emission percentage change input values for the table are based on those
      calculated in the letter from Chester France to Dr. Gary Whitten
      described in footnote 6.
      Commingling assumed half of that at higher market shares based on
      analysis in Appendix I.D.


      A 1.0 psi waiver would therefore have a large impact on the VOC emission
inventory.  EPA believes this would have a significant adverse impact on ozone
levels in the areas of the country already facing the worst ozone problems.
EPA does not agree with the arguments made by several commenters that for
reactivity and other reasons this increase in VOC emissions would not increase
ozone (see subsection D).   This increase in VOC emissions would also place a
serious obstacle in the way of the states in their planning to meet the Act's
reasonable further progress requirements.

      EPA would appear to have the flexibility, under certain circumstances,
to establish a different VOC or toxics emission reduction standard for ethanol
blends.   Under section 211(k)(1), EPA is to consider "cost..., any nonair-
quality and other air-quality impacts" in establishing requirements that
obtain the greatest achievable emission reductions from gasoline
reformulation.  This mandate, in conjunction with the minimum requirements of
section 211(k)(2)  and (3), arguably would authorize EPA to establish a less
stringent RVP or VOC performance standard for ethanol blends if the
circumstances warranted and the ethanol blends still met all the minimum
requirements of section 211(k), including the minimum VOC and toxics
reductions required under section 211(k)(3).   The Renewable Oxygenates Program
proposed by EPA in February 1993 was based in part on this approach.  However,
for all of the reasons described above and elsewhere, EPA does not believe
that it could take this approach in the context of a 1.0 psi waiver for
ethanol blends.  EPA believes that ethanol can and will play an important role
in reformulated gasoline without a 1.0 psi waiver, and that granting such a
waiver would therefore be unreasonable.

      With a 1.0 psi waiver,  the RVP standard for reformulated gasoline under
the first phase of the program would have to be roughly 7.1 psi in both
northern and southern areas if the ethanol blends were to still achieve the
minimum VOC and toxics reductions required under section 211(k).  This would
not be a reasonable exercise of EPA's discretion given the cost, energy, and
other impacts described elsewhere.  Even in the southern portions of the
country, where EPA does believe a 7.1 RVP standard is currently feasible, a
                                      11

-------
1.0 psi waiver for  ethanol  is not appropriate.  In the context  of  a  1.0 psi
waiver, the RVP standard under Phase I in southern areas  would  have  to be 6.2
psi if comparable emission  reductions are to be achieved  under  the
reformulated gasoline program in the South as in the North;  otherwise little
or no environmental benefit would be achieved in southern areas under the
reformulated gasoline program.  Furthermore, since the majority of the
southern reformulated gasoline required under the Act is  sold in California
which has its own strict requirements for reformulated gasoline, a waiver for
ethanol in southern areas would be of little or no benefit to the  ethanol
blending industry.

3.    Gasoline Volatility Rulemaking Decision

      Several commenters argued that EPA should allow a 1.0  psi waiver for
ethanol blends in the reformulated gasoline program for the  same reasons EPA
allowed such a waiver in its national regulations of gasoline RVP.   EPA
disagrees with these commenters for a number of reasons.

      EPA established limits on the summertime volatility of gasoline in two
Phases, pursuant to its broad regulatory authority under  section 211 (c) of the
Act.8   These  limits  applied nationwide.   For both Phase I  and Phase II RVP
regulations EPA set a 1.0 psi waiver for ethanol blends that met certain
alcohol content requirements.  The primary reason EPA adopted such a waiver
was to avoid the near catastrophic economic impact on that industry  if no such
waiver was allowed. EPA was convinced of the reality of  this jeopardy, and
the inability of the industry to survive it.  At that time there was no
significant alternate market for ethanol such as widespread  use of the ether
ETBE or a large-scale program to require the use of oxygenates  in  various
cities.  In that context, EPA believed it would require strong  evidence of
severe environmental consequences before it would impose  a policy  which might
eliminate that industry.  There was no evidence of such severe  environmental
consequences, and instead EPA's concerns over the negative air  quality impact
of 1.0 psi RVP waiver had been moderated during the rulemaking  process  (See 55
FR 23665, 23666).   At the same time, EPA announced that it would continue to
explore the economic and air quality issues involved with a  1 psi  RVP waiver,
and reserved the right to propose changes in the RVP regulations if
appropriate  (See 55 FR 23666) .

      EPA's concern over the environmental impact of a 1  psi RVP waiver
stemmed from the evidence showing that substantial VOC emission increases were
known to occur with ethanol blends relative to conventional  gasoline.  This
RVP and resulting VOC emission increase, however, existed with  ethanol blends
prior to the volatility control rule.  Thus, providing a  waiver under the
volatility control  rule required only the same RVP control from ethanol blends
as from non-ethanol containing gasoline.  Since the rule  was not expected to
increase the use of gasohol with its higher emission potential  (the  market
share of gasohol had been approximately 8% nationwide),  EPA  believed that
providing a waiver  would not significantly compromise the environmental
benefits of the program nationwide.  To illustrate this point,  the volatility
control rulemaking  reduced  the volatility of all gasoline by approximately 2.7
psi RVP.  The 1 psi RVP waiver for the 8% of the gasoline market represented
      8  Phase I  regulations  applied  to the  1989  through 1991
summertime control  period (52  FR 31274,  August 19,  1987),  while
Phase II  regulations applied  thereafter  (5 FR 23658,  June 11,
1990).  EPA recently revised  the Phase II regulations to
conform with  the requirements  of §  211(h),  added  in the Clean
Air Act Amendments  of 1990  (56 FR  64704  (December  12, 1991) .
                                     12

-------
by ethanol blends, thus, represented a loss of only 0.08 psi across all in-use
gasoline, or less than 3% of the volatility control that would otherwise have
been achieved by the rulemaking.  To the extent that ethanol blends achieved
lower exhaust VOC and CO emissions, these only served to offset what was
already considered to be a small loss in emission control relative to the
control achieved.  As a result, providing the waiver did not result in any
environmental detriment, but just slightly diminished the potential
environmental benefits that could otherwise have been achieved by the
regulations.

      In balancing the severe economic impact of not allowing a RVP waiver
with the generally modest concerns about the VOC and ozone impacts of such a
waiver, EPA reasonably exercised its broad discretion under section 211 (c) (1)
of the Act and provided a 1.0 psi RVP waiver in the national RVP regulations.

      The situation confronted in this rulemaking is significantly different.
First, the existence of a 1.0 psi waiver in the reformulated gasoline program
does not raise the same kind of life or death situation confronting the
ethanol industry in EPA's RVP rulemakings.  The requirements for reformulated
gasoline do not apply nationwide, thereby maintaining markets for ethanol in
conventional gasoline during the summertime.  In addition, refiners have no
choice on whether or not their reformulated gasoline will be oxygenated.
Based on the year-round oxygen content requirement for reformulated gasoline,
and the wintertime oxygen requirement for gasoline marketed in numerous CO
nonattainment areas, EPA believes that the demand for ethanol will increase
significantly without an RVP waiver, either as an oxygenate or as a blendstock
to produce ETBE.   This is in marked contrast to the situation confronting EPA
and the ethanol producing industry in the national RVP rulemaking.

      Furthermore, the environmental impact of an RVP waiver under the
reformulated gasoline rulemaking would be significantly more severe than it
was under the volatility control rulemaking.  As discussed above in subsection
C.2, a 1.0 RVP waiver coupled with the reformulated gasoline program's oxygen
requirement could result in a significant increase in the use of ethanol
during the summer months, leading to significant emission increases and the
elimination of a large percentage of the ozone-related benefits of the
program.  This stands in stark contrast to the less than 3% loss of the
emission reductions from the gasoline volatility control program as a result
of the waiver.

      Another significant difference between this rulemaking and the previous
national RVP rulemaking is the difference in statutory authority.  The
nationwide RVP regulations were based on EPA's broad discretionary authority
under section 211(c) of the Act.  An RVP or VOC performance waiver could be
established for ethanol blends in the reformulated gasoline program only if it
was authorized and justified under section 211(k).  That provision, however,
places many more constraints on EPA's discretion than section 211 (c),
including, for example, the minimum mandatory reductions in emissions of ozone
forming VOCs and toxics.  This and the other statutory provisions in section
211(k) limit EPA's discretion to allow a 1.0 psi waiver for ethanol blends.
For all the reasons described above and elsewhere, EPA does not believe that a
1.0 psi waiver for ethanol blends would be proper under section 211(k) of the
Act.

4.     RVP Increase From Ethanol Blending

      Some comments suggested that ethanol blends may not actually cause an
increase in the volatility of in-use gasoline.  Other comments suggested that
the volatility increase disappears at ethanol concentrations of 20%.   A great
deal of data and information collected by both EPA and outside parties on a
wide variety of gasolines demonstrates that ethanol does in fact increase the
vapor pressure of in-use gasolines when blended in at low concentrations.


                                      13

-------
Some of this data is  discussed  in  section G.  There is less data available as
to what the vapor pressure  increase  is at ethanol concentrations above 10
volume percent,  but  the  available  information suggests that blending as much
as roughly 50% ethanol  into gasoline  still  increases the vapor pressure of the
blend.9   Since blends of 10 volume percent ethanol or  less  are all  that  could
lawfully be used in  the  reformulated  gasoline program at the present time, the
fact that the volatility increase  disappears with blends with greater than 50%
ethanol is not relevant  to  this  rulemaking.  Despite the available data, if
ethanol were to be determined not  to  cause  an increase in the volatility of
the blend, then there would be  no  need for  a waiver since ethanol could easily
comply with the RVP  requirements.

      Some comments  were also received suggesting that the volatility increase
for ethanol blends was  due  to the  butane  in the gasoline, and not the ethanol
being added to the gasoline.  While butane  is a hydrocarbon found in gasoline
with a high vapor pressure,  and  as such represents the largest fraction of
evaporative emissions from  today's gasolines, the RVP increase which occurs
when ethanol is added to gasoline  is  not  due to butane.  Ethanol, being an
alcohol, is a polar  molecule that  has a low vapor pressure when relatively
pure due to a strong tendency to self-associate through hydrogen bonding.
When mixed with gasoline the hydrogen bonding is greatly reduced causing a
large increase in the vapor pressure  of the mixture.  The increased vapor
pressure increases the  tendency  of all of the hydrocarbons in the mixture to
evaporate, not just  butane.  The vapor pressure increase would occur even if
all butane were removed  from the gasoline prior to blending with ethanol.  In
fact, as the vapor pressure of  the base gasoline decreases (and butane content
decreases) the RVP increase resulting from  adding ethanol to the gasoline
increases rather than decreases.   In  any  case, the source of the volatility
increase in ethanol  blends  is not  important — no matter the source,  blending
ethanol into gasoline at the levels expected in the reformulated gasoline
program leads to significant increases in volatility.

5.    Mandate for Blendstock and Ethanol  Market Share

      In the context  of  the waiver discussion a number of commenters also
suggested that if a  waiver  were  not granted that EPA should promulgate
provisions to mandate that  refiners either produce a blendstock capable of
being blended with ethanol  downstream and still meet the criteria for
reformulated gasoline certification,  or use ethanol as the oxygenate in a
certain fraction of  the  RFG they produce.   In the alternative, one commenter
suggested providing  a 1.0 psi waiver  for  ethanol blends if refiners did not
provide an adequate  supply  of sub-RVP blendstock.  While these approaches
would ensure the use  of  ethanol  in the reformulated gasoline program during
the summer months, EPA  does not  believe that it has authority under either
§211(c)  or 211(k)  to  impose the  suggested provisions under the present
circumstances.   Furthermore, as  discussed in section G., these approaches
would not necessarily be environmentally  neutral even though no increase in
RVP would result.
      9  "Volatility  Characteristics  of Gasoline-Alcohol  and
Gasoline-Ether Fuel  Blends,"   Robert  L.  Furey, General Motors
Research  Laboratories,  SAE  Paper  No.  852116,  October  1995.

                                     14

-------
A.    Reactivity

      A number of  comments were received which attempted to find methods by
which the evaporative VOC emission increases associated with splash  blending
ethanol into gasoline could be offset.  These comments stated that while the
RVP increase resulting  from blending ethanol into gasoline may cause an
increase in evaporative VOC emissions, the additional exhaust VOC and CO
reductions resulting from the additional oxygen in a 10 volume percent ethanol
blend and the lower  relative reactivity of emissions from ethanol blends
offset any increase  in  ozone that might occur.  As a result,  they stated that
this justified either the RVP waiver discussed above, or the inclusion of
relative ozone reactivity in the determination of the VOC performance
standards for reformulated gasoline.  As discussed in the April 16,  1992
proposal, one method suggested for incorporating the ozone reactivity benefits
of ethanol blends  was to include carbon monoxide (CO) in the definition of
what comprises a reactive VOC and adjust its mass to take into account its
relative ozone reactivity.  This method was referred to as the "carbon mass
equivalent" method.

6.    Statutory Authority for Reactivity Based Adjustment

      The text of  section 211(k) does provide EPA with certain discretion to
consider the relative reactivity of different volatile organic compounds;
however, it is clear from this section that EPA may not consider reactivity
for purposes of compliance with the minimum emissions reductions required by
section 211(k)(3).   Section 211(k)(1) requires that EPA establish requirements
for reformulated gasoline that obtain the greatest achievable reductions in
emissions of "ozone  forming volatile organic compounds... and toxic  air
pollutants" and the  provision for certification of reformulated gasoline again
refers to ozone forming volatile organic compounds (see section 211(k)(4)(B)).
While the statute  defines toxic air pollutants (see section 211(k)(10)(C)), it
defines neither ozone forming volatile organic compounds nor volatile organic
compounds.  The use  of  the undefined phrase ozone forming volatile organic
compound would,  therefore, appear to provide EPA with certain discretion to
consider ozone forming  potential or reactivity of different VOCs.

      At the same  time, Congress clearly limited this discretion by  explicitly
stating that the minimum reductions in emissions of ozone forming VOCs and
toxic air pollutants required under section 211(k)(3) "shall be on a mass
basis."10  Since both of these provisions must be given meaning, there would
appear to be only  one reasonable interpretation.   EPA would have discretion to
consider reactivity  in  deciding what volatile organic compounds should be
considered ozone forming, but compliance with section 211(k)(3)(B) Ts minimum
percentage reduction in ozone forming VOCs would be based on mass reductions,
without further adjustment for ozone forming potential.  In effect,  EPA would
establish what VOCs  are ozone forming VOCs and reformulated gasoline would
then have to meet  at least a minimum percentage reduction,  on a mass basis, of
these VOCs.  Since the  limitation to reductions on a mass basis refers to
requirements under section 211(k)(3)(B), if EPA required greater than these
minimum emissions  reductions, then its authority under section 211(k)(1) would
appear to give EPA the  discretion to consider reactivity in requiring these
      10   In a  similar fashion Congress,  in  the separate
provision addressing anti-dumping,  established limits  on the
average  per gallon  emissions  of volatile organic  compounds,
without  reference to ozone forming  potential,  and again
specified that compliance with the  anti-dumping requirements
was  to be measured  "on  a  mass basis."   See  §211(k)(8)(A)  and
(C) .

                                     15

-------
further reductions, as long as  the minimum required reductions were  obtained
on a mass  basis.

      The  legislative history for this provision  supports EPA's
interpretation.  The bills passed by the House and Senate both contained
provisions for reformulation of gasoline, requiring reductions in "ozone-
forming volatile organic compounds" and "ozone-forming potential of  volatile
organic compounds,"11 respectively.  Both provisions required 15% reductions
in these volatile organic compounds, but neither  provision referred  to mass
emissions.   In fact, the House  bill specifically  required that the
Administrator consider "reactivity and relative toxicity" in establishing the
reformulated gasoline requirements.  In conference, Congress removed the
reference  to reactivity and relative toxicity found in the House bill, and
added the  provision calling for measurement of emissions on a mass basis.  The
House bill's anti-dumping provision was generally adopted by the conference
committee,  but again a provision for determining  compliance on a mass basis
was added.   Although there is no indication in the conference committee report
on why these changes were made, the clear import  of these changes is that
Congress intended to restrict the Administrator's ability to consider
reactivity,  at least in the context of the minimum reductions required in
Phase I and II of the reformulated gasoline program.  The statements of
various congressmen during debate of this provision support this view,
indicating the conscious choice of a mass basis standard.12

      EPA's interpretation is also reasonable when considered in relation to
another important ozone control provision in the  Act.  Section 182 requires
that various ozone nonattainment areas submit SIP revisions providing for a
schedule of percentage VOC reductions from a 1990 baseline of actual VOC
emissions.   Compliance with this requirement is based on reductions  in the
mass of the VOCs, without adjustment for reactivity.  Interpreting section
211(k)  to  allow for significant increases in VOC  emissions would present a
major impediment for state compliance with the section 182 requirements.
Commenters failed to provide a  resolution to this problem other than suggest
that states be provided with a  credit for the VOC increases caused by ethanol
blends. EPA does not have authority to grant such a credit.

      Several commenters also suggested that EPA  define CO as an ozone forming
VOC.   While the technical and policy reasons for  rejecting this approach are
discussed  later, it is also important to note that EPA recently conducted a
rulemaking to define volatile organic compounds for a wide variety of CAA
purposes.   In that final rule EPA followed its prior regulatory approach and
excluded CO from the definition, as well as several other chemical compounds
      "Section  217, S.  1630  as passed by  the Senate on January
23,  1990.

      12   See, e.g.  136  Cong.  Rec.  S16922  (daily  ed. October  27,
1990)  (statement  of Senator  Durenberger)  ("The  VOC reduction
requirement is  15  percent in the year 1995.   The  reduction  is
measured  in the mass of  emissions comparing emissions  from
[baseline  vehicles burning baseline  gasoline]  and the  same
vehicles  burning  any proposed reformulated gasoline....");  and
136  Cong.  Rec.  H12900  (daily ed. October  26,  1990)  (statement
of Representative  Hall)  ("...  the use of  a mass basis  test...
compares  the percentage  difference between emissions  from
baseline  gas and  from  the proposed reformulated gas"
(discussing the need to  control the  vehicle variable  in
evaluating a reformulated gasoline performance)).

                                    16

-------
that include carbon.  The agency clearly stated that CO and certain other
compounds were well understood by the scientific community not to be
considered as organic and, therefore, could not be considered as a VOC (see 57
FR 3944, February 3, 1992).  Furthermore, as discussed below the relative
reactivity of CO is very low such that even if it were considered to be
organic, it would be excluded from the definition of VOC on this basis as well
under the established EPA regulations.  The comments do not present material
to change this view, and it would be arbitrary for EPA to do so now, given the
significant, adverse impact this would have on EPA's and the states' overall
ozone control strategy.

7.    Technical Validity

      a.  CO as an Ozone Forming VOC

      As noted above, EPA has discretion to determine what VOCs are "ozone
forming" for purposes of section 211(k).   The low relative reactivity of
methane and more recently ethane was the basis for their exclusion from the
Agency's definition of VOC in the rulemaking discussed in the previous
paragraph, and ozone forming VOC in the most recent proposal for this
rulemaking.  According to the California Air Resources Board's (CARB)  relative
reactivity scales, methane is more than one order of magnitude less reactive
than the next least reactive hydrocarbon, ethane, and more than two orders of
magnitude less reactive than the average hydrocarbons contained in gasoline
vehicle emissions.  Since there are a number of compounds in the emissions
from gasoline vehicles that are just slightly more reactive than ethane,  there
is no justification at this time for excluding any additional compounds from
the current definition.  Ethanol's relative reactivity is more than five times
that of ethane, while methyl tertiary butyl ether's  (MTBE) is 2-1/2 times that
of ethane and half that of ethanol.   In fact, 50 of the 168 hydrocarbons in
CARB's scale have ozone reactivities less than that of ethanol.  Furthermore,
since the emissions from ethanol blends are comprised of essentially the same
compounds as the emissions from MTBE and pure gasoline blends  (with the
exception of the oxygenates themselves),  just in different amounts, excluding
any of the compounds with the next lower relative ozone reactivity would
effect the performance of ethanol blends little relative to other fuels.

      Just as there is no justification for excluding any additional compounds
from the definition of what comprises reactive VOCs, there is also no
justification for including any additional compounds at the present time.  A
number of comments were received suggesting EPA should include CO in the
definition of ozone forming VOC.  CO, however has a relative ozone reactivity
of roughly one-fifth that of ethane and nearly one-fiftieth that of average
gasoline hydrocarbon emissions.  Given its low reactivity relative to the
majority of hydrocarbons, it would be inappropriate to include it in the
definition of ozone forming VOC.  This is particularly true given the relative
magnitude of CO emissions to hydrocarbon emissions in the exhaust of gasoline
vehicles.  If the Agency included CO in the definition of VOC the relative
ease of reducing CO emissions to hydrocarbon emissions through fuel
modification would turn the reformulated gasoline program into a CO control
program and virtually eliminate all ozone benefits.

      b.  Reactivity Adjustment

      Even if the Clean Air Act provisions allowed the minimum VOC performance
standard to be met on an ozone reactivity weighted basis, the science is far
from adequate to support a move away from mass based standards.  All reactive
VOCs eventually react to form ozone.  Their relative reactivity is merely a
measure of how quickly they react compared to other VOCs.  Depending on the
ambient conditions, some VOCs may not react before being blown out of the non-
attainment area by the prevailing winds.   However, they will then react
downwind to form ozone.  Thus, even the less reactive VOCs can represent a


                                      17

-------
significant concern,  particularly in  areas such as  the northeast  corridor,
where there are many  ozone non-attainment areas at  some distance  downwind of
each other.  Furthermore, the determination of the  relative reactivity of the
various VOCs is a function of the ambient conditions.  As a result, the
relative reactivity of various VOCs may be very different from day to day and
from city to city making it difficult to apply one  set of relative reactivity
assumptions nationwide.  A compound which demonstrates a low tendency to form
ozone in one city may demonstrate a higher tendency in another city, or even
the same city on a different day.

      Even if we were to consider reactivity in the context of the
reformulated gasoline rulemaking despite the legal  constraints and scientific
uncertainties, it would not be beneficial to ethanol.  An analysis using the
GARB relative reactivities in a letter from Chester France to Dr. Gary
Whitten13 indicates that an ethanol blend with 1 psi higher RVP would, after
adjustments for ozone reactivity, increase VOC emissions from a typical 1990
model car by approximately 19% relative to other reformulated gasolines (if
the commingling effect of ethanol blends with other gasolines is  incorporated
into the analysis).

      Several commenters submitted modeling analyses prepared by  Systems
Applications International (SAI)  as justification for much larger ozone
reactivity benefits for ethanol blends than estimated by EPA.  EPA has
carefully reviewed SAI's analyses and rejects their conclusions because the
analyses are fraught  with invalid assumptions and inconsistencies which make
it impossible to use  the studies to appropriately quantify the effect on ozone
resulting from the use of a reformulated gasoline containing ethanol with a
1.0 psi RVP waiver relative to a reformulated gasoline containing MTBE.  EPA's
comments and criticisms of these studies are contained in various letters and
memorandums located in the Docket and in Appendix I. C.14 15 16  17 18  In the most
recent study for the  Council of Great Lakes Governors, SAI attempted to
correct many of the problems which they recognized  existed with the earlier
      13    See  Footnote 6

      14    Letter from Chester J. France, Director,  RDSD  to Dr.
Gary Whitten,  Chief  Scientist, Systems Applications
International, September  24,  1993.

      15    Letter from Paul A.  Machiele, Fuel Studies and
Standards Branch,  to Dr.  Gary Whitten, Chief Scientist,  Systems
Applications  International,  September  8,  1992.

      16    Letter from Paul A.  Machiele, Fuel Studies and
Standards Branch,  to Eric Vaughn,  President, Renewable  Fuels
Association,  August  6, 1992.

      17    Letter from Charles L. Gray Jr.,  Director, Regulatory
Programs and  Technology,  to  Timothy McNulty, Executive
Director, Council  of Great Lakes Governors,  June  4, 1993.

      18    Evaluation of the  Ethanol Air Quality Study by the
Council of  Great Lakes Governors,"  Memorandum from Paul  A.
Machiele to Richard  A. Rykowski, Chief, Fuel Studies and
Standards Branch,  and Chester J. France,  Director,  Regulation
Development and Support Division,  August  5,  1993.

                                   18

-------
studies.  However, a number of serious errors were made in the protocol
causing the study's results to be of little scientific value.  One of the most
serious of these errors being the exhaust VOC and CO emission credit assigned
to the ethanol by comparing an ethanol blend with 3.5 weight percent (wt%)
oxygen to an MTBE blend with 2.0 wt% oxygen.  This comparison, assumed in all
three of these SAI studies, is invalid for analyzing the air quality impacts
of fuels under the reformulated gasoline program due to the oxygen averaging
provisions of the Act.  Furthermore, the magnitude of the credit given for the
additional oxygen is significantly larger than that now predicted by EPA's
Complex Model.  Due to this and other erroneous assumptions, the study
drastically underpredicted the evaporative emission increases and
overpredicted the exhaust emission benefits with ethanol blends relative to
MTBE blends (not only in highway vehicles, but even more so in non-road
engines which are not even considered under section 211(k)  of the Act)  for a
reformulated gasoline scenario.  Since the study assumed only a small
contribution of gasoline vehicles to the total VOC inventory, the result of
the combined set of assumptions was very little increase in VOC emissions for
the ethanol blend.  The study also assumed large NOx emission increases for
both on-road and non-road engines with ethanol blends which, combined with the
very low VOC/NOx ratio assumptions caused the urban ozone decreases shown with
the ethanol blend despite a small increase in overall VOC.   Such an assumption
with respect to NOx is in violation of the requirements for reformulated
gasoline under section 211(k), and is also inconsistent with the currently
available information on the effect of fuels on NOx.

      Even if the studies performed of Chicago and New York by SAI were valid
studies, it would be difficult for EPA to justify using their results to
reactivity weight VOC emissions in the reformulated gasoline rulemaking.
Reformulated gasoline will be sold in many different areas of the country,
representing a wide variety of ambient conditions.  Given our current
understanding of air chemistry, widely different results would be expected for
different cities, episodes, and timeframes.  Thus, it would be difficult to
draw conclusions that are applicable nationwide from urban airshed modeling of
just one or two ozone episodes in one or two cities.  This is especially the
case when the model is used to focus on the ozone effects of one specific
change, such as the use of ethanol, in motor vehicle fuel.   The model
considers the entire emissions inventory  (mobile, point, area and biogenic
sources) for the area, as well as a large number of other variables for which
input assumptions must be made.  All of these factors can influence the
results and mask the impact of fuel changes on overall VOC mass emissions and
ozone.  The influence of these factors is one of the primary reasons for the
widely different results of studies conducted by SAI for the Illinois Corn
Growers Association, National Corn Growers Association, and Council of Great
Lakes Governors and studies conducted by SAI for the Auto/Oil Air Quality
Improvement Research Program and in the past for EPA.

8.     Carbon Mass Equivalent

      In the April 16, 1992 proposal, EPA asked for comments on a concept
referred to as the carbon mass equivalent  (CME)  method, a modified approach
for treating CO as a VOC.  Rather than merely counting the mass of CO
emissions as VOC as discussed above, this approach would exclude the mass of
oxygen from CO and apply certain other adjustments to its mass before counting
it as VOC.  EPA received a number of comments supporting such an approach,  as
well as a number of comments rejecting such an approach.

      EPA does not believe the CME approach represents a valid approach in the
context of our reformulated gasoline rulemaking.  Excluding only the oxygen
mass from carbon monoxide would still greatly overemphasize the ozone forming
potential of carbon monoxide relative to other gasoline vehicle emissions by
roughly a factor of 20.  As discussed above, this would quickly turn the
reformulated gasoline program into a CO control program and virtually


                                      19

-------
eliminate all ozone benefits.  The only rationale for making additional
adjustments to the mass of CO would be its relative ozone reactivity.  This
would improperly incorporate ozone reactivity into the rulemaking, contrary to
EPA's authority under section 211(k).   Furthermore, there is no justification
for giving credit for only the CO emission reductions resulting from one type
of oxygenate as was suggested by some of the comments, or even just oxygenates
as a group.  The effects on CO of all fuel modifications would have to be
determined and appropriate credit given.  If the proper adjustment were made
to the carbon mass to reflect the relative reactivity of CO and other
oxygenates and fuel parameters that affected CO emissions were also provided
this carbon mass equivalent, then ethanol blends would receive little
additional credit relative to other reformulated gasolines.


A.    Benefits of Ethanol to the Nation

      A number of comments were received supporting a strong role for ethanol
blends in reformulated gasoline through either an RVP waiver,  former President
Bush's ethanol announcement or some other means on the basis of the benefits
of ethanol to our Nation relative to a situation where ethanol is excluded
from participation in the reformulated gasoline market.  Such stated benefits
included:  improved energy security,  increased rural development, reduced
foreign trade deficit, reduced U.S. unemployment, reduced federal agricultural
subsidies, and many others.

      EPA believes that the ethanol industry is important to the nation, and
provides many valuable benefits.  As stated before, EPA does not believe that
ethanol will be locked out of the reformulated gasoline program, or that the
demand for ethanol will shrink.  Instead, EPA believes that ethanol will play
a large role in providing oxygenate to meet the increased demand for oxygen
from the year-round reformulated gasoline program and from the wintertime CO
programs.  In that context, the issue before the Agency is to identify and
quantify the benefits to the nation from any additional increase in demand
over and above that anticipated from these programs that might result from
either a one psi waiver under reformulated gasoline or from an incentive
program such as the program proposed by EPA in the February 26, 1993 proposal.
EPA, of course, would also need to look at any adverse impacts from such an
increase in demand, as well as its authority to allow a one psi waiver or an
ethanol incentive plan like that proposed in February, 1993.  Viewed in that
light, many of the comments submitted in support of the ethanol industry
position provide little help in resolving these issues.  Few of the comments
contained any analysis supporting how the claimed benefits would be achieved,
and when analysis was provided, it focused on comparing the benefits of a
large increase in the demand for ethanol with a large decrease in demand.
None of the comments showed why EPA's April 1992 proposal would fail to
achieve the claimed benefits.  In addition, none of the comments provided a
direct link between such benefits and special provisions for ethanol.
Finally, none of the comments provided an analysis showing that the claimed
increase in benefits was justified in light of the cost and environmental
impacts from such a program.

      As discussed in Section A., EPA believes ethanol is not excluded from
the reformulated gasoline market.  Rather, EPA believes that the reformulated
gasoline program will increase the market share of ethanol over and above the
increase already achieved under the wintertime oxygenated fuels program.  As a
result, EPA believes that if the benefits to the nation discussed in the
comments will occur as a result of increased ethanol demand, then the
incremental increase in ethanol demand resulting from the reformulated
gasoline program will result in a corresponding incremental increase in such
benefits to the nation.  As discussed in section C., however,  further
encouraging the use of ethanol through an RVP waiver could easily eliminate
all of the VOC emission reductions of the reformulated gasoline program.  As


                                      20

-------
such,  there would not  appear to be any rationale for attempting to  increase
the ethanol market in  this manner.

      Similarly,  the loss in environmental control discussed in Section G.
resulting from the renewable oxygenate provisions of the February 26,  1993
proposal strongly argue  against its use to expand the market share  of  ethanol
despite the potential  incremental benefits to the nation that might otherwise
accrue.  Increasing the  ethanol market to just 30% under the proposed  program
is estimated to eliminate roughly 40-50% of the VOC emission reductions of the
reformulated gasoline  program  (as discussed below in Section G.).   This is a
large environmental cost, particularly since it is possible that ethanol
market shares of 30% may occur without such provisions.   Furthermore,  comments
submitted by the Department of Energy indicate that there in fact would be no
energy or crude oil benefits under an ethanol incentive  program as  proposed by
EPA.19  The marginal net  energy benefit of producing ethanol from corn  using
the current processes  is offset by the increased energy  consumption at the
refinery to offset the RVP increase resulting from the use of ethanol.  This
belief that there are  no energy benefits was echoed by various other
commenters based on their own refinery modeling.  Given  the potential  lack of
energy benefits,  and the dramatic environmental impacts  of either a waiver as
discussed above,  or the  special incentives in the February 26,  1993 proposal
as discussed below, EPA  does not believe that the reformulated gasoline
rulemaking is the appropriate program by which benefits  to the country such as
those described in the comments should be obtained.


A.    Feasibility of Renewable Oxygenate Proposal

      In the February  26, 1993 proposal, EPA proposed an incentive  program to
encourage the use of renewable oxygenates in reformulated gasoline. This
program, in keeping with the October 1, 1992 announcement by former President
Bush,  essentially provided an RVP incentive of 0.3 psi RVP in a refiner's
performance standard for the use of 30% renewable oxygenates in his RFG.  The
VOC emission increase  resulting from the ethanol and other renewable oxygenate
blends would be offset by a reduction in the emissions from non-ethanol
blends.  In order to provide additional incentive for a  refiner to  actually
use the renewable oxygenate, however, EPA proposed that  refiners would forfeit
the ability to obtain  the RVP incentive to their competitors if they did not
commit to use the renewable oxygenate.  Without such an  added incentive EPA
feared some refiners might decline to use the renewable  oxygenate and  instead
opt for the performance  standard 0.3 psi more stringent.  EPA also  proposed
various tracking and recordkeeping provisions to ensure  that the program
resulted in the required RVP control  (and VOC control under the complex model)
in all areas covered by  the reformulated gasoline program.

      A large majority of comments from parties on both  sides of the ethanol
issue claimed that EPA's proposal was unworkable and too intrusive  to  permit
efficient market operation.  Ethanol producers and blenders and agriculture
interests argued that  the program was so burdensome and  complex and the
limitation on the ethanol market share which received the incentive so
restrictive that any incentive the program was designed  to create was
      19   As discussed previously,  neither the 1.0  psi waiver
under section  211(h), adjustments  to reactivity,  nor
adjustments based  on CO  emission  reductions  can be used to
offset  the RVP and resulting  VOC  increase in ethanol blends.
An  energy analysis based on such  assumptions would therefore
not be  relevant.

                                     21

-------
effectively eliminated.  Instead they argued for an outright waiver from the
RVP requirement as discussed above.

      Similarly, the oil industry argued that the fuel tracking and
recordkeeping requirements on each batch of fuel were not feasible, especially
given the short lead time between now and the programs initiation in 1995 to
satisfy the proposals new tankage and distribution requirements.  The
increased complexity in production, transportation, distribution, and
enforcement of the program would dramatically increase the cost of the
reformulated gasoline program to the consumer.  Such a cost increase was not
justified given the small, or even negative energy benefits of the program  (as
discussed above),  and the negative environmental impact of the program  (as
discussed below).   Both the oil and non-ethanol oxygenate industries argued
that ethanol and its derivative ethers would already represent a considerable
portion of the RFG market and did not need any additional subsidies.
Providing incentives for ethanol and its derivative ethers would place other
competing oxygenates at a considerable economic disadvantage with a resulting
economic impact on the industries which currently produce these oxygenates.
They argued that any additional incentives or federal subsidies for ethanol
should be considered in forums other than the RFG rulemaking,  as section
211(k)  did not authorize EPA to provide such a incentive for ethanol use.

      EPA acknowledges the burden caused by the additional fuel tracking,
recordkeeping, and enforcement provisions of the February 26,  1993 proposal.
While EPA believes that the program's design provides a large economic
incentive for the use of renewable oxygenates such as ethanol in reformulated
gasoline relative to other oxygenates (refer to Section VIII.  of the Draft
Regulatory Impact Analysis),  EPA also acknowledges the considerable increase
in cost for fuel production,  transportation, distribution, and enforcement
which resulted from such an intrusion into the marketplace which may offset
much or all of this incentive.  Given the significant increase in the cost and
burden to the regulated industry, the various questions raised in regard to
its workability and impact, the lack of any energy benefits (as discussed in
section E.), and the significant environmental loss  (as discussed in section
G.), EPA does not believe it to be appropriate, in light of the provisions of
section 211(k), to adopt the renewable oxygenate program proposed in February,
1993.  The lack of support from the ethanol industry for this proposal further
indicates that adoption of this proposal would not be appropriate.

      In response to the criticisms of the proposal, EPA investigated a number
of options aimed at making the program more workable by reducing the fuel
tracking, recordkeeping, and enforcement burden associated with the proposal.
While such options tended to make the program more workable from the
standpoint of the refining and fuel distribution processes, they also tended
to either reduce the assurance that the environmental benefits of the program
would be achieved in all areas covered by the RFG program, or to place
additional restrictions on the RVP and emission performance of the ethanol
blends (thereby reducing the certainty that ethanol would actually be used).
Furthermore, they did little to address the environmental concerns discussed
in Section G.  Given this and the other concerns with the proposal  (cost, lack
of energy benefits, significant environmental loss, etc.), EPA did not believe
these options to be appropriate for promulgation either.

      A number of comments were received concerning the justification and
rationale for various detailed provisions of the renewable oxygenate incentive
proposal (e.g., the percent of ethanol in the fuel needed to receive the RVP
incentive).  Given the fact that we are not promulgating the proposal, EPA
does not believe there to be any reason to evaluate the relative merits of
these provisions.
                                      22

-------
A.    Emission Impacts of the Renewable Oxygenate  Proposal

      As discussed  above, granting ethanol a 1.0 psi  waiver from the
reformulated gasoline requirements would result  in a  significant loss (if not
a total loss)  in  the environmental control of the  reformulated gasoline
program.  EPA also  received a number of comments that there are significant
negative environmental impacts from providing incentives  for ethanol use as
under the February  26, 1993 proposal.  Despite EPA's  best efforts to propose
an environmentally  neutral incentive program for renewable oxygenates, VOC
emissions would be  expected to rise under the program for four main reasons;
fuel commingling, distillation effects, unrestricted  early use of the complex
model, and greater  permeation of oxygenates through non-metal fuel lines and
hoses.  Appendix  I.D. contains EPA's analysis of the  ethanol commingling
effect.  References located in the docket20 discuss its impact on VOC
emissions.  EPA's analysis of the distillation effects are discussed in detail
in a recent paper published in the technical literature21, and again, the
references cited  above discuss its impact on VOC emissions.  The emission
impact of unrestricted early use of the complex model is  discussed below.  EPA
has not had an opportunity to evaluate the effects of oxygenates on fuel line
and fuel tank permeation, but the reader is referred  to SAE paper number
920163 suggesting dramatic increases in permeation emissions with oxygenates,
particularly alcohols.22

9.    VOC Emission  Increase Estimate

      Assuming a  30% market share for ethanol in reformulated gasoline during
the summer months,  the commingling effect is estimated to result in roughly a
2-2.5% increase in  fleetwide  (6-8% if applied to ethanol  market share only)
gasoline vehicle VOC emissions (for an assumed 0.15-0.20  psi increase in in-
use RVP), the distillation effect a 3% increase  in gasoline vehicle emissions
(relative to CAA  baseline gasoline), and unrestricted early use of the complex
model a 1-2% increase, for a total increase of 6-7.5% relative to a scenario
with no gasoline  reformulated with ethanol.  This  represents a loss of 40-50%
of the VOC emission control required under section 211(k).  EPA is in the
process of collecting additional data and information which will enable a
better analysis of  the distillation effect, but does  not  expect the current
emission effect estimate to change appreciably once the new data is
incorporated due  to the scientific theory which  supports  the effect.
Inclusion of fuel tank and fuel line permeation, however, would be expected to
further increase  the estimate.

      Comments were received suggesting EPA take into consideration additional
effects on emissions when analyzing the in-use impact of  ethanol blends such
as fuel weathering  and enrichment of the air fuel  ratio when vehicles
accelerate above  and beyond that which occurs during  EPA's certification test
      20
            Refer to  footnotes 5  and 6
      21   "Running  Loss Emissions From Gasoline  Fueled Motor
Vehicles," C.E.  Lindhjem  and D.  Korotney,  U.S.  EPA,  SAE  Paper
No.
931991,  October,   1993.

      22   "Fuel-Alcohol Permeation Rates  of Flouroelastomers,
Fluoroplastics,  and  other Fuel  Resistant  Materials,"  W.M.  Stahl
and   R.D.  Stevens, E.I. du Pont  de  Nemours and  Co.,  Inc.,  SAE
Paper No.  920163, February,  1992.

                                     23

-------
procedure.  Fuel weathering is already contained in the MOBILE model and as
such already reflected in EPA's evaporative emission estimates,  including the
estimates of emission increases resulting from commingling.   The emission
differences between EPA's certification test procedure and those which occur
during in-use driving, often referred to as off-cycle emissions, is currently
the focus of a great deal of study.  However, there is not adequate data and
information available at the present time to approximate with any confidence
the magnitude of such emission effects.23  In addition, enrichment of exhaust
emissions during acceleration is only one of the off-cycle emission impacts
currently being evaluated.  It is possible that off-cycle vehicle driving
patterns will have just as large of an effect on non-exhaust emissions as
exhaust emissions.  Due to the lack of data, however, EPA is not able at this
time to determine whether such off-cycle emissions improve or exacerbate the
emission impacts of ethanol blends described above.

10.   Unrestricted Early Use of CM

      EPA proposed three different options for early use of the complex model,
that is, optional use of the complex model before its use became mandatory.
The first option would require that the reformulated gasoline achieve VOC,
toxics, and NOx performance that was no worse than the complex model would
achieve for a simple model fuel using that refiner's 1990 baseline for sulfur,
T-90 (E300),  and olefins.  This was designed to ensure that early use of the
complex model would not upset the expected emissions benefits in the early
years of the reformulated gasoline program.  For example, if a refiner could
optionally use the complex model and measure performance against the CAA
baseline, then those refiners with individual baselines that were more
stringent than the CAA baseline (for example, lower sulfur levels)  could be
expected to use the complex model early and take credit for their lower sulfur
levels.  However, those refiners who had baselines that were less than the CAA
baseline, for example with sulfur levels higher than the CAA baseline, would
be expected to use the simple model as use of the complex model would require
that they offset the emissions impact of higher sulfur levels.  Unrestricted
early use of the complex model might in effect unbalance the emissions balance
imposed under the simple model which established refiner-specific caps on
sulfur, T-90, and olefins.  EPA discussed the pros and cons of this option in
the April 1992 proposal.

      EPA also proposed a second option in the April 1992 proposal, where
refiners could use the complex model early for fuel distributed to VOC control
region 1  (southern areas of the Country),  and measure performance against the
CAA baseline and not their individual baselines.  To limit interference with
the enforcement scenario proposed by EPA,  a cap would be placed on maximum RVP
levels.

      A third option for early use of complex model was proposed in February
1993 as part of the incentive program for ethanol blends.  Under this option,
performance under the complex model would be measured using either the CAA
baseline or the refiners' 1990 baseline, at their option.  This was described
as providing the maximum flexibility for refiners, and, therefore,  the maximum
incentive for ethanol use.  However, EPA was still concerned this would lead
to a reduction in the emissions benefits expected from the simple model.

      Unfortunately, it is difficult to assess with any precision the
environmental impact that would result from unrestricted early use of the
complex model.  EPA does not have any information at the present time to
estimate the number of refiners that would opt for early use of the complex
model,  nor does EPA have any information on the number of refiners with 1990
baseline levels of sulfur, T90, and/or olefins below the CAA baseline levels,
      23     Refer to footnotes 5  and  6
                                      24

-------
or what the levels of these parameters are for these refiners.  As shown in
Table 1-2, if refiners had low levels of sulfur, T90 (high E300), and/or
olefins, they would achieve substantially lower emissions reductions than
required by the simple model.  Assuming a normal distribution about the CAA
baseline 7/8 of all refiners would have a low baseline of one or more of
sulfur, T90, and olefins.  Given these estimates, and the incentive for low
baseline refiners to opt for early use of the complex model relative to the
CAA baseline gasoline, it would not seem unreasonable to conclude that the
proposal to allow such an option would sacrifice 1-2 percentage points of the
VOC emission reductions expected under the simple model.

Table 1-2  Emission Impacts of Unrestricted Complex Model Use
Fuel
Parameter
Sulfur (ppm)
E300 (%)
Olefins
(vol%)
VOC %Change
From
Baseline*
NOx Change
From
Baseline*
CAA
Baseline
Level
339
83
9.2
N/A
N/A
Refiner
A
150
83
9.2
-1.6
-5.4
Refiner
B
339
92
9.2
-1.4
+ 0.8
Refiner
C
339
83
3 .0
-0.7
-0.5
Refiner
D
150
92
3 .0
-2.3
-5.2
      Based on the Phase I Complex Model
A.
Reg Neg
      A number of comments were received suggesting that the renewable
oxygenate provisions of the February 26, 1993 proposal violated the agreement
reached through regulatory negotiation.  As a result,  these comments suggested
that if EPA were to promulgate the provisions, EPA would face numerous
lawsuits.  While EPA shares the desire of the commenters to maintain the
provisions reached through negotiation, it must be made clear that the
negotiated agreement is not in and of itself legally binding.  EPA's final
rule must be based on the authority provided by the Act.  The regulatory
negotiation process is intended to develop a consensus among the affected
parties on how the Act should be interpreted and implemented.  To EPA's
knowledge, there was no consensus reached during the negotiation on any
special provisions or waivers for ethanol blends.  Only if a deviation from
the negotiated agreement, such as the renewable oxygenate provisions,  also
resulted in a deviation from EPA's legal authority under the Act, however,
would EPA be at risk in any ensuing litigation.
                                      25

-------
A.    Appendices

Appendix I.A.     Ethanol Sales in 1990

      The following table shows and analysis of the ethanol sales in 1990 in
those areas currently covered by the reformulated gasoline program.   The
gasohol sales data on a state by state basis were obtained from the  1992
National Petroleum News Factbook.  Sales data for each covered area  was
obtained by ratioing the gasohol sales in the state by the proportion of the
state population represented by the covered area based upon 1990 census data.
This same population weighting is what EPA used to obtain the percent of
national gasoline consumption in each covered area.
                                      26

-------
Insert ethanol sales table here
                                      27

-------
Appendix I.B.      Relative  Economics of Oxygenates in RFG

      Historically,  oxygenates  (primarily ethanol) have been used as fuel
extenders or more recently  as octane enhancers.  With the advent of the
oxygenated fuels program to control wintertime carbon monoxide emissions  and
the reformulated gasoline program to control summer ozone, oxygenates now have
value for other purposes, as well.

      First, the addition of an oxygenate to gasoline impacts the Reid vapor
pressure (RVP)  of the  blend.  Since Phase I reformulated gasoline (RFG) has
restrictions on RVP, and Phase II RFG requires significant reductions in  VOC
emissions that  will  likely  require further reductions in RVP, the impact  of
oxygenates on the RVP  of the blend is an important factor to consider.

      Second,  oxygenates are high in octane, and thus enhance the octane  of
the finished gasoline.   Since other means of providing octane, such as the
addition of aromatics,  are  costly and often have negative consequences for
emissions,  oxygenates  are valuable for their role as octane enhancers.

      Finally,  in the  context of an oxygen content requirement such as exists
in the RFG requirement  for  2.1 wt% oxygen (on average), oxygenates are valued
for their oxygen content.

      The relative costs of three major oxygenates, ethanol, MTBE, and ETBE,
have been compared based on their purchase prices and their values as fuel
extenders,  RVP  modifiers, octane enhancers,  and oxygen sources.   Since MTBE  is
believed to be  the oxygenate likely to be used in reformulated gasoline in the
largest volumes (given  the  current market conditions),  the costs of ETBE  and
ethanol have been compared  relative to the cost of MTBE.24

      The purchase price of the oxygenates was determined from the spot market
prices or reasonable assumptions about their costs.  The price of MTBE was
estimated to be $0.70  per gallon, based on the recent  (1993) spot prices
reported in trade press publications such as Octane Week and Oxy-Fuel News.
The price of ethanol was assumed to be $1.20 per gallon, the typical price
over much of 1993.   (See below for further discussion of the price of
ethanol.)  The  purchase price of ETBE  ($1.01/gallon)  was estimated from the
price of MTBE,  assuming ethanol is purchased at $1.20 per gallon and the
capital and operating  costs for ETBE production are slightly higher (5% and
10%, respectively)  than for MTBE since somewhat large equipment  is necessary
to process the  same  volume  of isobutylene into ETBE.   The cost per gallon of
isobutylene was assumed to  be equivalent for both ethers.

      The value of the  oxygenates as a fuel extender was based on the recent
spot market price of unleaded regular gasoline at the U.S. gulf  coast (around
$0.50/gallon).   Since  the analysis was performed on the basis of the relative
cost per gallon of each oxygenate, the oxygenates have equivalent values  as
fuel extenders  and thus the magnitude of the value assumed for the gasoline
displaced is unimportant.

      The values of  the oxygenates as octane enhancers, oxygen additives,  and
RVP modifiers were estimated based on market trends and on the results of the
refinery modelling used to  develop costs for Phase II RFG.  It was assumed
      24ETBE  like MTBE  can be  added at  the refinery.   While  it
shows promise  for  the  future,  ETBE has  not  been widely used in
the  market  to  date.   Ethanol  blends cannot  travel  through
petroleum pipelines  and thus  ethanol  is blended at the
distribution terminal.

                                     28

-------
that the oxygenates were worth $0.006 per octane number added per gallon of
blend produced.  All of the oxygenates have octane ratings higher than the
base gasoline, so all have a value as octane enhancers.  The value of the
oxygen provided was assumed to be $0.003 per weight percent oxygen added per
gallon of blend produced.  Since ethanol has the highest oxygen content per
gallon, it has the highest value as a source of oxygen.  The cost of reducing
the RVP of the blendstock  (to maintain a constant RVP in the final blend) was
estimated for this analysis to be $0.005 per psi per gallon of blend produced.
(Recent refinery modeling, however, suggests that in some cases the cost of
RVP control could be considerably less expensive.)  Ethanol has a higher RVP
than gasoline, so ethanol incurs a cost because the RVP of the blendstock must
be lowered.  The ethers have lower RVPs than gasoline and thus are valuable
because the RVP of the blendstock does not have to be reduced to accommodate
these oxygenates .

      Ethanol is eligible for a $0.54/gallon tax credit which adds to its
value.  In this analysis, it was assumed that ethanol used in ETBE was
similarly eligible for this tax credit.  Ethanol is also eligible for
additional tax credits and other forms of economic support as high as
$0.50/gallon of ethanol in a number of States.  These credits have not been
factored into this analysis.

      Table 1 presents the comparison of the oxygenates.  As the table shows,
ethanol costs less and ETBE costs more than MTBE for use as a gasoline
oxygenate.  When expressed on a cost per gallon of finished gasoline basis,
the ethanol has a small advantage over MTBE of around 0.4 cents per gallon
gasoline.  ETBE, however, is slightly more costly than MTBE, adding an
additional 0.8 cents per gallon gasoline.  Distribution costs have not been
included in this analysis.  Since ethanol cannot be shipped through pipelines,
it tends to incur greater distribution costs if shipped over long distances.
Distribution costs will vary significantly depending on the locations
involved, but for many of the RFG areas the incremental cost for distribution
of ethanol should be less than 5 cents/gallon of ethanol.  ETBE, however, is
likely to be produced in the Gulf, while the ethanol is largely produced in
the Midwest.  Therefore, ETBE will also incur an additional cost for the
distribution of the ethanol to the Gulf, which could reach about $0.01 per
gallon of ETBE produced.

       Table 1: Costs of Oxygenates at 2.1 wt% Oxygen  ($/gal oxygenate)
                         Ethanol Price:  $1.20/gallon

Purchase Price
Fuel Extender Value
RVP Modifier Value
Octane Enhancer Value
Oxygen Value
Blender's Tax Credit*
Total Cost ($/gal oxygenate)
Total Cost ($/gal RFG)
Ethanol
1.20
-0.50
0.050
-0.160
-0.108
-0.54
-0.06
-0.004
ETBE
1.01
-0.50
-0.020
-0.148
-0.049
-0.23
0.06
0.008
MTBE
0.70
-0.50
0.001
-0.141
-0.056
0.00
0.00
0.000
      Federal credit only, individual state credits have not been included in
      this analysis
                                      29

-------
      Recently, the price of ethanol has decreased as the oxy-fuels season
started and refiners found that there was sufficient oxygenate supply.   The
price reported for the October-November timeframe has been around $1.05 per
gallon.  If this price is assumed for ethanol,  the relative values ethanol and
ETBE as oxygenates compared to MTBE change somewhat, as shown in Table  2.   At
the lower purchase price, combined with the tax credit, ethanol costs $0.01
less per gallon of gasoline than MTBE.  ETBE costs the same as MTBE on  this
basis.

       Table 2: Costs of Oxygenates at 2.1 wt%  Oxygen  ($/gal oxygenate)
                         Ethanol Price:  $1.05/gallon

Purchase Price
Fuel Extender Value
RVP Modifier Value
Octane Enhancer Value
Oxygen Value
Blender's Tax Credit*
Total Cost ($/gal oxygenate)
Total Cost ($/gal RFC)
Ethanol
1.05
-0.50
0.050
-0.160
-0.108
-0.54
-0.21
-0.013
ETBE
0.94
-0.50
-0.020
-0.148
-0.049
-0.23
0.00
0.000
MTBE
0.70
-0.50
0.001
-0 . 141
-0.056
0.00
0.00
0.000
      Federal credit only, individual state credits have not been included in
      this analysis
                                      30

-------
Appendix I.e.     Critique of the Draft Final Report:  "Comparison of the Air
                  Quality Effects of Ethanol and MTBE in Reformulated Gasoline
                  in the New York Region in 1995," for National Corn Growers
                  Association, by Gary Z. Whitten, Systems Applications
                  International, August 18, 1992.


General Overview of Comments:

      As discussed in the August 6, 1992 letter to Eric Vaughn shown in
Appendix I.e., EPA has serious reservations with respect to the ability of the
UAM to adequately evaluate the relative ozone impacts of oxygenated blends in
the New York airshed.  Past experience with New York indicate that transport
of pollutants into the airshed dominates the result and makes establishment of
appropriate initial and boundary conditions extremely difficult.  As a result,
we recommended that some other city be modeled instead where transport was not
such an overwhelming issue.  Instead,  SAI proposed a new approach of adjusting
the initial and boundary conditions based on the results from the first day.
It was agreed, however, that external  peer review of this method would be
obtained by a panel of recognized experts in the field.  It was also agreed
that a number of sensitivity analyses  would be performed to allow evaluation
of the study's results.  These agreements were not fulfilled.  The results
cited in the introduction and conclusion sections of the final draft report
contained in the docket to this rulemaking are not based upon any runs for
which the transport issue was addressed.  Furthermore, while a sensitivity
analysis was performed using an approach for dealing with transport, peer
review of the method was not obtained, nor was enough information provided in
the draft final report by which EPA could evaluate the approach.  As a result,
EPA has no choice other than to maintain our position expressed in the August
6, 1992 letter to Eric Vaughn that the results from the modeling of New York
are not valid.

      Furthermore, our preliminary review of the report provided suggests that
even if the transport issue were adequately addressed, the inappropriate
assumptions made in the protocol would invalidate the results.  Other than
using UAM-IV instead of UAM-V and including the specific air chemistry for
ethanol and MTBE, the protocol for New York is virtually identical to the
protocol used in the Chicago study.  As a result, most of the comments and
criticisms of that study contained in  Appendix I.B also apply to this study.
Comments specific to the New York study are discussed below.  As with the
previous Chicago study, however, the documentation which was provided for both
the protocol and the final results does not allow for a thorough evaluation of
the appropriateness of the study or usefulness of its results.

      The EKMA modeling which was performed for 14 cities also appears to be
of limited usefulness as discussed in  the comments below.

Detailed Comments on the New York UAM:

1)    Model Validation

•     It is necessary to show that the model is adequately predicting
      historical ambient ozone  (and other pollutant)  levels for the ozone
      episode being modeled.  There was no discussion of this.  Presumably
      since the same model and episode were used by AQIRP, it has been
      validated, but it is not known if the same initial and background
      assumptions were used.

•     The modeling, with the exception of some sensitivity runs (which were
      downplayed and ignored in the final conclusions) apparently did not
      address the impact of transport  on the initial and boundary conditions.
                                      31

-------
•     In the sensitivity runs that were performed,  the method of adjusting the
      boundary conditions for transport was not adequately explained.   While
      it may be directionally better than making no adjustment,  based on the
      limited explanation in the final report,  the method used appeared to be
      of little merit.  It used the effects of ethanol on the inventory 55 km
      downwind from New York to adjust the boundary conditions since New York
      is 55 km downwind from Philadelphia.  As a result,  it appears to ignore
      the impacts of Newark, Jersey City, Baltimore,  Washington D.C.,  Trenton,
      etc.   It is also unclear how well it represents the upwind effects of
      Philadelphia since no explanation was given as  to what the time period
      was over which the emission effects of ethanol  were evaluated in order
      to adjust the initial and boundary conditions.   The difference would be
      dramatic between evening, early morning,  or afternoon emission effects.
      Furthermore, as in the Chicago study, the primary compounds of interest
      in this study (ethanol and MTBE) were substituted with similar CBM-IV
      species in the air chemistry for the boundary conditions.

2)     Episode Selected

•     The maximum temperature, diurnal temperature swing, and any other
      relevant episode information were not provided.  Without such
      information it is difficult to evaluate the results.

•     There was no explanation provided for the assumptions made for the
      initial and boundary conditions  (other than the adjustments made for the
      sensitivity runs).

3)     Selection of 1995 as the Future Study Year

•     The ozone impact of ethanol and MTBE blends in  the year 1995 was
      evaluated.  The ozone impact of ethanol in particular, is the sum of a
      number of impacts,  some positive and some negative.  One of the positive
      impacts is the likelihood that ethanol provides additional exhaust VOC
      and CO emissions beyond that provided by 11 volume percent MTBE.
      However, these benefits are greater for older vehicles than for recent
      model year vehicles.  Likewise, the evaporative emission detriments of
      ethanol are greater for vehicles not designed to meet EPA's upcoming
      enhanced evaporative and running loss emission  control.

      The choice of 1995 as the focus of the study (the earliest possible year
      to evaluate reformulated gasoline)  maximizes the exhaust emission
      benefits of ethanol.  However, assuming the same assumptions were made
      for New York as for Chicago, unrealistic and insupportable assumptions
      were made concerning the implementation of new vehicle evaporative
      emission controls in 1995.

      Since the decision to grant or not to grant ethanol a 1.0 psi waiver
      will affect reformulated gasoline well into the next century, evaluating
      the impact only in 1995 is unacceptable.   At minimum, evaluations should
      be performed in 2000 and 2005, as well.

4)     Inventory

•     The final report gives no detail of the control assumptions for area and
      point sources assumed for 1995.  If as in the Chicago study, no controls
      were assumed for these sources despite the application of stringent
      mobile source controls, and rapid growth was instead assumed, the
      relative importance to ozone levels of the increases in mobile source
      emissions resulting from the use of ethanol blends is underestimated.

*     As confirmed in subsequent conversations and discussions since the time
      of the previous Chicago study's release,  the MOBILE model used in SAI's


                                      32

-------
work was not EPA's MOBILE4.1 model, but a version of MOBILE4.0 modified
by SAI and/or DOE.  The impact of this on the results of the UAM,
however, may be small based on the analysis performed by SAI in a
September 10 memorandum from JGH to Dr. Whitten.  The original MOBILE
model used by SAI with assumptions of enhanced I/M and a new evaporative
test procedure and Tier I controls gave similar results as using the
actual MOBILE4.1 model with corrected I/M assumptions, no new
evaporative test procedure assumptions, and Tier I controls.  The
MOBILE4.1 results, however, do not appear to be consistent with the
actual inventories used.  Apparently there are other assumptions made in
developing the inventories from the MOBILE model outputs that change the
relative contribution of the various pollutants and emission types.
Until EPA is provided with such information we cannot ascertain the
validity of the mobile source inventory.

Use of the same evaporative test procedure assumption as in the Chicago
study is just as inappropriate in the New York study  (if in fact the
same assumption was made).  Either 1998 or a later year should have been
modeled  (decreasing the exhaust emission benefit of the ethanol blend),
or no evaporative emission reduction should have been claimed for 1995
due to the new test procedure.

If the same assumption was made here as in the Chicago study, the
inspection and maintenance program assumptions are unrealistic.  Not
only are these assumptions inconsistent with the proposed I/M rule, but
they also provide an effectiveness which can not be achieved in-use.
Full effectiveness is not likely for a few years after the start of the
program which is currently scheduled for the beginning of 1995.  Once
again, either 1998 or a later year should be modeled, or reduced
benefits claimed in 1995.

Presumably, the same assumptions were used for the exhaust VOC emission
benefits for the oxygenated blends as in the Chicago study.  As
discussed in the comments on the Chicago study, the assumption for
ethanol is too high relative to that for MTBE, especially considering
the I/M scenario assumed in the study.  At a meeting in Ann Arbor on
9/11/92, apparently Dr. Whitten agreed with EPA's assessment of the
oxygenate effects as he raised no objection to EPA's assumptions.  As a
result, SAI overstates the exhaust VOC benefit of ethanol blends in the
inventory.  This is particularly important since much of the air quality
benefit given to ethanol to offset the evaporative and running loss
emission increases was in the form of exhaust VOC emissions.

The fuel assumption of ethanol at 3.5 wt% oxygen and MTBE at 2.0 wt%
oxygen is not valid for a reformulated gasoline scenario.  The oxygen
averaging provisions under the reformulated gasoline program will limit
the maximum average oxygen content in the fuel to 2.1 wt% oxygen.  When
combined with the minimum oxygen content requirement for 1.5 wt% oxygen,
if ethanol is used at 3.5 wt% oxygen, it will only be used in 30% of the
fuel, with the remainder of the fuel containing ethanol or some other
oxygenate at 1.5 wt% oxygen.  If the study is to have any relevance to
the reformulated gasoline rulemaking as it purports, then both the
ethanol and MTBE scenarios should have the same overall oxygen content.
This will essentially eliminate the exhaust VOC and CO emission credits
applied to ethanol in the study, and greatly increase the ozone
estimates for the ethanol scenario relative to the MTBE scenario.

Presumably the non-road inventory was based upon EPA's non-road study.
However, there is no explanation of this.  Furthermore, there is no
explanation made for the temporal, spacial, and growth assumptions made
for non-road emissions.
                                33

-------
Presumably the non-road inventory in Table 1 is only the gasoline
portion of the non-road inventory.  However, it is not possible to know
this from the information provided.  Neither the non-road or the diesel
related inventory is broken down into its constituent pieces allowing
for any clarification.

One glaring error in the New York study which benefits ethanol
tremendously is that it was assumed that there were no evaporative or
running loss emissions from non-road vehicles.  Since non-road vehicles
and engines do not have any evaporative emission control systems, the
non-road evaporative VOC inventory should be significant.  As a result,
SAI provides a very large exhaust VOC credit to ethanol relative to MTBE
(roughly half that of the exhaust credit from on-road vehicles) and
ignores completely the evaporative emission increases with ethanol
relative to MTBE.  EPA acknowledges that there is a general lack of
information on the evaporative VOC emissions from non-road sources, and
we will be working to improve our understanding of their magnitude.
However, to ignore them completely while including the exhaust inventory
when the fuel effect being evaluated has dramatically different effects
on exhaust and evaporative emissions is unconscionable.  Either non-road
emissions should have been assumed constant for both scenarios, or some
sensitivity runs performed with reasonable assumptions made as to their
magnitude.

For the effect of oxygenates on non-road exhaust emissions, SAI used
data from Southwest Research Institute on one 4-stroke engine and one 2-
stroke engine.  This test data may be of some use for demonstrating
certain trends, but is hardly adequate to quantify the effect of
oxygenates on exhaust emissions from non-road vehicles.  The fuels used
in the testing are not representative of in-use fuels, and the two
engines are not representative of the wide variety of non-road emission
sources.  As a result, there should be no distinction made between the
MTBE and ethanol fuels for the effect of oxygenates on non-road emission
sources in this study.  The results of the study may be too sensitive to
any assumptions made for non-road, which cannot be substantiated.

Why is there only a 9.8% increase in refueling emissions for LDVs with
ethanol, and why is it different for MDVs and HDVs?  Based on the vapor
pressures of the fuels in question, there should be roughly a 12-13%
increase.

If the same assumption was made here as in the Chicago study, Stage II
refueling controls were assumed by SAI to reach full benefit by 1993,
but probably will not be in effect until 1995 and will not reach full
effectiveness for several years.  Once again, either 1998 or a later
year should have been modeled, or reduced benefits claimed in 1995.

Based on the protocol, the same data, information, and assumptions were
supposedly made in the New York study as in the Chicago study regarding
the impact of ethanol and MTBE on VOC, CO, and NOx emissions.  This held
true in the New York inventory for CO emissions. The percent changes in
CO emissions from MTBE to ethanol in the inventory were exactly the same
as in the Chicago study.  However, the percent changes for VOC and NOx
were different.  Exhaust VOC emissions decreased with ethanol by smaller
percentages, NOx emissions increased by smaller percentages  (except for
HDVs which now showed a decrease in NOx instead of an increase),  running
loss emissions increased by smaller percentages, and evaporative
emissions increased by slightly larger percentages.  What is the
justification and rationale for these changes?  Temperature conditions
were likely different prompting different evaporative and running loss
emission changes, but the changes should have been in the same
direction.  Furthermore, given that the changes in CO were the same,


                                34

-------
      they should have been for exhaust VOC and NOx as well.   There especially
      does not appear to be any justification for why NOx was now assumed to
      decrease with ethanol for heavy-duty vehicles.

      The study assumes 100% replacement of gasoline  with gasohol under the
      ethanol scenario.  This is unrealistically high and results in the
      commingling effects of ethanol on in-use RVP being ignored.

      Presumably SAI made the same faulty VMT and ambient temperature
      adjustments to the running loss emissions that  were made in the Chicago
      study.   The VMT adjustment shifts more emissions to the morning hours
      than is appropriate based upon EPA information.  The temperature
      adjustment then zeros out most of these inflated morning running loss
      emissions and forces the remaining running loss inventory to the end of
      the day when there is not adequate time for it  to react and contribute
      to peak ozone.  Emissions do not vary linearly  with temperature so a
      linear weighting of running loss emissions with ambient temperature does
      not reflect the diurnal variability of running  loss emissions.  For
      example, the methodology used assumes running loss emissions are zero at
      the morning low temperature, eliminating virtually all  running loss
      emissions from the morning commute.  While running loss emissions are
      temperature dependent, all currently available  data suggests that
      significant running loss emissions occur even at morning low ambient
      temperatures down to about 40'F.  The effect of these VMT and
      temperature assumptions together is to drastically reduce the impact of
      running loss emissions on ozone formation.  As  a result, the largest
      source of emission increases (running losses)  in the inventory with
      ethanol blends is virtually eliminated.

      SAI could have very easily adjusted the running loss emission profile
      for temperature in an appropriate manner by running the MOBILE model for
      each hour of the ozone episode.  In addition to being erroneous,
      however, the adjustment made to running loss emissions  based upon the
      time of day is in and of itself biased in favor of ethanol.  To make
      adjustments only for running loss emissions without also making
      adjustments for all other portions of the inventory is  inappropriate.

      Another serious oversight in the study is that  while fuel effects on
      refueling emissions were taken into account, for some reason the fuel
      effects on storage emissions which represent a  three times larger
      portion of the inventory were ignored.   The effect of this is to ignore
      some of the adverse effects of volatility increases with the ethanol
      blend.

      The methodology used to develop the emission inventories (and speciation
      profiles) for medium and heavy-duty vehicles was not explained in the
      protocol.  Based on follow up discussions, these inventories were based
      on crude extrapolations of light-duty vehicle data which are not
      supportable.  The overall effect on the results of these assumptions is
      unknown, but it does raise additional uncertainty as to the validity of
      the study's results.
5)     Speciation
      Refer to comments in Appendix I.E.  on the Chicago study.   Apparently the
      same speciation profiles were assumed for both studies.   As for that
      study,  EPA cannot agree with the profiles developed and recommends the
      use of our own speciation profiles.

      Formaldehyde and acetaldehyde emission increases with ethanol are
      ignored, while the formaldehyde and isobutylene increases with MTBE are
      not.  This is inappropriate.


                                      35

-------
*     Using EPA default species for non-road is inappropriate since they are
      based on light-duty vehicle data.  The SwRI report from which the
      oxygenate effects data was taken suggests that 2-stroke exhaust is
      predominantly unburned fuel,  and that 4-stroke exhaust resembles that of
      conventional gasoline engines.  If your assumption is true that 70% of
      non-road exhaust is 2-stroke, then the speciation profiles should
      approximate those of the raw fuel.

•     The CO reactivity in a 1988 SAI study was assumed to be 71% greater than
      that of ethane.  The Carter reactivity for CO is l/5th that of ethane.
      What was used in these studies?  It could have a significant impact on
      the results.


6)     Results

*     Results discussion ignores first day results.  New York gets blown
      through in a day.  As a result first day results are just as valid as
      second day.  Much higher - why?

•     The results are difficult to interpret since isoline results for the
      ethanol blend scenario and no gasoline mobile source scenario were not
      provided separately.

•     The adjustment for transport in the sensitivity runs resulted in ozone
      increases.  These were ignored and downplayed in the discussion of the
      results.

*     As in the Chicago study, the results discussion still focussed only on
      the peak ozone and not the larger and geographically dispersed ozone
      increases.

*     The sensitivity of ozone to VOC emission increases from ethanol were
      more appropriate than in the Chicago study, but still lower than for
      other studies - including the SAI/AQIRP study of New York.  Since New
      York was the least sensitive city in the AQIRP work, larger increases in
      ozone would be expected in other cities.


Detailed Comments on the EKMA Modeling of 14 cities

•     Mobile emission inventories for the cities were based on the New York
      UAM inventory and were not specific to the different cities.  Not
      specific for temperatures or RVPs,  etc.  As a result, they were not
      appropriate.

*     As discussed above and in the comments on the Chicago UAM study, the New
      York inventory has numerous inaccuracies and is based on many
      unrealistic assumptions.  As a result, the inventories used for the EKMA
      modeling are also invalid.

•     The baseline gasoline inventory was arbitrarily assumed to be MTBE
      blended gasoline.

*     Again, default chemistry is used for the many different VOC compounds,
      and not the specific chemistry for the compounds being evaluated in the
      study.

•     No explanation is given why for Philadelphia and San Diego the EPA
      inputs could not be used, and instead why the precursor ratios had to be
      raised and the default chemistry replaced with an "updated version".
                                      36

-------
Given the results for these two cities showed that removing all mobile
source emission increased ozone, something appears to have gone awry.

No explanation given for the city selection and its representativeness
to the nation or the RFC program.

It is not possible from the results to say that in 4 cities "ethanol
reduces ozone compared to the MTBE base."  The results for three of
those cities should have been thrown out since eliminating all mobile
source emissions increased ozone.

The results showed a -0.7 to 4.1% increase in the mobile source
contribution to ozone with the use of ethanol (average of 1.75%) .   While
much smaller than predicted by other studies, this is nevertheless a
very significant increase, representing a large fraction of the ozone
benefits achieved by RFG.

How can the 23.2% motor vehicle contribution to the VOC inventory in
Chicago be reconciled with the 5.1% assumed in the Chicago UAM study?
                                37

-------
Appendix I.D.      Commingling Effect of Ethanol


I.     INTRODUCTION

      Ethanol  has a Reid Vapor Pressure (RVP)  of 2.4 psi, and most gasolines
will have an RVP of 6.5 to 9.0 during the summer months.  When ethanol is
mixed with the gasoline a non-linear increase in the RVP is observed.  For
example, if gasoline with an original RVP of 8.0 is mixed with ethanol (10% by
volume)  the resultant RVP is approximately 9.1.  This RVP increase can be
offset at the  refinery by using a lower RVP gasoline so when the two
substances are mixed together the desired final RVP can be achieved.  However,
due to the non-linear nature of ethanol's blending RVP, the mixing of ethanol
blends with other non-ethanol containing gasolines downstream of the refinery
in vehicle fuel tanks can result in an additional vapor pressure increase
across the in-use pool of gasoline.  This RVP increase caused by fuel mixing
is what is known as the commingling effect.  This study will only address the
commingling effect in vehicle fuel tanks.   Other effects outside of the
vehicle fuel tanks are assumed to be small due to fuel segregation through the
fuel distribution system.

      The commingling effect is not a simple matter to model for any
geographical area.  Many factors must be taken into account in order to
accurately describe the overall consumer refueling behavior which is so
critical in establishing a viable commingling model.  Such factors are the
refueling patterns of the consumers using the various gasoline blends, their
loyalties to various brands, the base RVP of the fuels used, the market share
of gasolines using ethanol as an additive, and whether the geographical area
in question is utilizing reformulated gasoline or not  (due to the cosolvent
effects of other oxygenates).  Such complexities and variables are extremely
difficult to account for using simple hand or spreadsheet calculations, thus a
computer program was developed to attempt to include as many of these
variables as possible in order to develop a more accurate estimate of the
commingling effects that ethanol as a gasoline additive has on in-use RVP.
                                      38

-------
II.   THE COMMINGLING MODEL


      A.    General Overview

      The commingling model developed here simulates  real  world conditions  by
performing 100 refueling events for each of 1000  vehicle owners,  for  a  total
of  100,000  refueling events.   The model utilizes  brand loyalty curves and
refueling pattern curves in order to determine the  quantity and type  of fuel
for each refueling event.

      The model is flexible enough to be used for a number of  different in-use
assumptions (a listing of  the computer program is contained in the  appendix).
One of these is whether it is a reformulated gasoline scenario where  the
ethanol blend is assumed to be  mixed with a MTBE  blend, or a conventional
gasoline scenario where the ethanol blend is mixed  with pure gasoline.   The
model, however, is not flexible enough to model a scenario where the  ethanol
blend is mixed with both MTBE blended gasoline and  conventional gasoline.
Another in-use assumption  of the model is that the  base RVP is the  same for
all blends.  This means that the RVP of the gasoline  that  is initially  mixed
with the pure ethanol is lower  such that all gasoline blends have the same
RVP.  Thus, the results from this commingling analysis are incremental  to any
RVP increase that might result  from RVP waivers for ethanol blends.   A  third
in-use assumption is that  of the ethanol market share in the particular region
being modeled.
      Two other important  variables in the model  are  the brand loyalty  and  the
consumer refueling pattern.  Due to a degree of uncertainty in the  available
for brand loyalty and consumer  refueling data, the  model was designed to allow
for a number of different  assumptions.  This was  done in order to bracket the
range of possible in-use conditions.


      B.    Brand Loyalty

      Figures #1 - #3 represent the various forms of  owner brand loyalty
curves that were derived for use in this computer model.   Figure #1 defines
the customer loyalty to a  particular brand of fuel  based on data collected  by
Arco.   This data, submitted to EPA in 1981 as part of ARCO's  oxinol  waiver
request, only has a few points  and indicated a great  propensity towards
extremely high owner loyalty25.   Of particular importance with  this  data is
the seemingly large percentage  of customers that  have 100% brand loyalty.   The
Arco data appeared to be unrealistic because of this  fact.  Much of the
reasoning behind the apparent skewing of this graph towards high loyalty may
be in the manner that the  data  was collected.  The  questionnaire only made  an
inquiry as to how many customers purchase    the  same brand of gasoline 75% to
100% of the time.  Thus it is the sensitivity of  the  data  collection  procedure
that produced this high loyalty skew.  One would  expect there  to be actually
very few customers who are 100% loyal to one brand  (never  use  a different
brand of gasoline, even on long trips), however,  this data serves well  as an
upper bound for consumer brand  loyalty. Two other curves were  developed in  an
attempt to exhibit consumer brand loyalty in a less dramatic measure.

      Figure #2 shows these two other curves.  The  curve labeled "Curve #2" is
an attempt to retain the general trend of the Arco  data yet adjust  for  the
concept that there would be a decrease in the percentage of owners  with high
loyalties  (80% to 100%)  to a particular brand. Curve #3 shows an assumed
normal distribution of loyalty  that assumes that  most people in major urban
      25U.S.  Environmental  Protection Agency,  Arco  Oxinol Waiver
Request,  1981.

                                      39

-------
areas  purchase fuel from a number (e.g. 2-6) different brands.   As  such, the
maximum percentage of  the time that  they purchase the brand they use most
often  tends to be rather low.  A fourth brand loyalty not shown in  Figure #2
but permitted in the model assumes that the consumers    all have zero
loyalty26.  This assumption is on the opposite end of the spectrum as the Arco
data and serves as the lower end of  the loyalty bracket.

      Figure #3 shows  each of these  brand loyalty assumptions as a  cumulative
percentage of the total.  The Arco data  (at the far  right in Figure #3) shows
the greatest trend toward high levels  of consumer loyalty,  while the other
three  curves show lesser degrees of  loyalty.

      EPA has recently obtained some additional brand loyalty data  through
contract with the NPD  Group that serves to support the assumptions  of curve
#227.   These results are shown in Table #1
      26Zero  loyalty assumes that  customers have  no preference  at
all  to the  brand of  fuel  they purchase.   Thus in these  cases
there is a  completely random determination of the fuel  brand
being used  with each fill-up of  fuel.

      27NPD Group  Inc., "Special Gasoline Analysis, Annual 1992
Data Prepared for: U.S.  Environmental Protection Agency".  1993.

                                   40

-------
Table #1 Brand Loyalty Data, Total U.S. Industry - MPD Group
Brand Grouping
Always use One Brand
Use 2 or 3 Brands
Use Many Different Brands
Percent Loyalty
37.8%
51.2%
11.0%
       By assuming that the heading "Always use one brand" indicates an actual
consumer loyalty of 90% - 100%,  the "Use 2 or 3 brands" indicates  a loyalty of
40% -  60%, and the "Use many different brands" corresponds to an actual
loyalty of 20% to 40%, then an average consumer loyalty of this  data can be
compared to that for the other brand loyalty curves.28  Using these
assumptions  consumer loyalty for the NPD data is  calculated to  range from 57%
to 73% (ave is 65%).   The average brand loyalties  for each of the  curve of
Figure #3 are: 75% for the Arco data, 64.5% for the adjusted Arco  data  (Curve
2) and 50% for the even distribution (Curve 3). Thus Curve #2 appears to
closely match the interpretation of the NPD data,  and may be the best
representation of consumer brand loyalty available for this model.

      There is, however, no one loyalty curve that will fit all  geographical
areas. One may expect large cities to have a lower overall owner  loyalty
because of the large number of different brands available in close proximity
to one another.  Conversely, rural areas should have a higher degree of
consumer loyalty, particularly if there is only one or two gasoline stations
available and thus only one or two brands.

      There are two assumptions made in this model that are concerned with
consumer loyalty.  The first assumption is that the loyalty curves the model
uses are applicable only to a fuel brand and nota  particular oxygenate.
There  was no data available on oxygenate loyalty of which EPA was  aware.
Another assumption of the model is that it does not allow for individual
loyalties for various grades of gasoline.     Ethanol tends to be  used more in
mid and premium grades of gasoline than in the regular grade. At  this time
EPA can not speculate on what effect, if any, these assumptions  may have on
the model results.  As more information becomes available that sheds light on
these  two points EPA may consider their applicability to this model.


      C.    Refueling Patterns
      28The  95% to  100% loyalty  indicates  that even though
consumers may report 100%  loyalty  only one purchase of  a
different brand per year is  enough to drop that  loyalty to
about 97% to 99%.   Therefore,  consumer response  to "always use
one  brand"  may not  indicate  their  actual  buying  patterns.   The
other loyalties,  45% to 50%,  and 25% to  30% represent an
assumed distribution among multiple brand purchases.  These
ranges were purposely made slightly higher than  a strict
division of 100%  by the number of  brands  involved to account
for  the likely preferential  use of one brand out of the total
number of brands  purchased.   Without any  further information on
the  loyalty to one  particular brand, when more than one brand
is purchased,  it  is difficult to establish unequivocally what
the  actual  brand  loyalty should be in the multiple brand
purchase scenarios.

                                   41

-------
      Two curves can be chosen in the model  in order to describe the  fuel fill
patterns of  each customer for each individual fill-up as a percentage of the
total tank capacity.  The data of Figure  #4  indicates the fuel fill level
prior to a fill-up.  The bars labeled "Case  1  (Actual Data)" are taken from a
General Motors  refueling survey of over  1100 refueling events29.  This data
shows a relatively uneven distribution in some areas.  The bars labeled "Case
2 (Assumed Fit)" are a smoothing of the  GM data.  This is an attempt  to
eliminate some  of the unevenness in the  data that may have been caused by
anomalies in the method of reporting the  fuel tank fill levels, particularly
at the 0.2 gallons in/tank capacity level on the graph.  It is also important
to note that whatever the fuel fill scenario that is used for any particular
run the model will always assume an existing "heel" in the tank of 10 percent
of capacity.  Thus, any numbers derived  from this graph are assuming  that the
tank capacity does not include the "heel".

      Figure #5 describes the possible levels of the tank after fill-up.  This
particular graph is for the case where the initial amount in the tank before
refill was equal to zero (except for the  "heel").  There are other fill
scenarios based on each possible pre-fill level of the tank (i.e.  0.1,  0.2,
etc.).  All  scenarios are based on the same basic information.  In Figure #5
the bars labeled "Case #1 (Actual Data)"  are again from the GM refueling
survey, while the bars labeled "Case #2  (Assumed Fit)" are a smoothing of
portions of  this GM data (particularly at the 0.8, 0.9, and 1.0 fill  levels).

      The reasoning behind this smoothing is that the original GM data had an
inordinate amount of data points representing a complete fill-up.   EPA
believes that the incidences of actual total fill-up are less than this due to
the manner in which the data was collected.  The general inaccuracy of most
gasoline fill level gauges,  many of which read full even though the tank is
much less than  full causes an over estimation of "full" reportings.   EPA hopes
to obtain more  substantive information to either support the present  data or
provide for  a more accurate representation of typical consumer fill up
patterns in  the future.


      D.    Vapor Pressure Curves

      In order  to assess the RVP boost that will be experienced with  the
mixing of an ethanol fuel with either a  non-reformulated or a reformulated30
gasoline a series of tests were performed to measure the resultant RVP boost
at various levels of ethanol/gasoline in  the case of non-reformulated gasoline
and ethanol/MTBE/gasoline in the case of  reformulated gasoline.  Tables #2 and
#3 show the  percentage of concentration  of the various constituents in the
tests for non-reformulated and reformulated gasoline, respectively.   The tests
in Table #3  adjusted the ethanol and MTBE concentrations to simulate  the
mixing that  would be experienced in an actual reformulated gasoline scenario.
The results  of  these tests were used to  produce the vapor pressure curves for
use in the commingling model.  Smoothed  graphs of the resultant curves are
shown in Figure #6.
      29U.S. Environmental  Protection  Agency,  "Draft Regulatory
Impact Analysis,  Control  of Gasoline  Volatility  and Evaporative
Hydrocarbon Emissions  from New Motor  Vehicles",  July,  1987.

      30  In this  computer program  reformulated gasoline  is
assumed  to be oxygenated  with MTBE.   All properties of  the
reformulated gasoline  and the non-reformulated gasoline were
assumed  to be identical except for the difference  in RVP boost
responses  when  mixed with another fuel containing  ethanol.

                                    42

-------
      The RVP boost represented in Figure #6  is only applicable at  the base
RVP that these particular tests were performed31.   In order to determine the
RVP boost at different base  RVPs a RVP adjustment  factor was determined.  This
factor is applied to the RVP boost in order to arrive at the final  adjusted
RVP boost of the mixture.

      The RVP adjustment factor was determined based on experiments performed
at General Motors in which ethanol was added  to gasolines of varying RVP32.
It was determined upon examination of the trends in this data that  the RVP
boost changed roughly 0.05 psi for every 1.0  RVP increase or decrease in the
base RVP.  The resultant straight line curve  that  was used in the commingling
model is shown in Figure #7.

      Although, in actuality this 0.05 psi RVP adjustment factor is at best an
estimate based on available  information, its  influence is actually  quite small
for the ranges of RVP that are generally being  considered (RVP 7.0 - 9.0).
Table #4 provides some of the numerical data  provided by GM upon which this
0.05 psi RVP adjustment was  based.

Table #2 - Non-reformulated Gasoline Testing Chart
Run III Percent of 10% II Percent of Base II Total Percent of
Number III Ethanol Blend in II Gasoline in Total II Ethanol* in
III Total Mixture || Mixture || Mixture
1A
2A
3A
4A
5A
6A
7A
8A
9A
10A
11A
12A
13A
14A
0 .0
2 .5
5.0
7.5
10 .0
15.0
20 .0
25.0
30 .0
40 .0
50 .0
60 .0
80 .0
100 .0
100 .0
97.5
95.0
92 .5
90 .0
85.0
80 .0
75.0
70 .0
60 .0
50 .0
40 .0
20 .0
0 .0
0 .0
0 .25
0 .5
0 .75
1.0
1.5
2 .0
2 .5
3 .0
4 .0
5.0
6 .0
8 .0
10 .0
 * Denatured without methanol
8.4.
31 These particular tests were  done  using  a base  RVP of
 It  has been found that  as the  base  RVPs become  greater
there is a  decrease in  the RVP boost  for  the same  percentage  of
added ethanol and conversely  at increasingly lower RVPs  there
is  an increase in the RVP boost greater than that  observed in
Figure #6.

      32General Motors  Corp.,  Letter  to U.S.  Environmental
Protection  Agency,  May  21, 1993.
                                    43

-------
Table #3  - Reformulated Gasoline Testing Chart
Run
Number
IB
2B
3B
4B
5B
6B
7B
8B
9B
10B
11B
12B
13B
14B
Percent of 10% II Percent of 11% II Total Percent of
Ethanol Blend in II MTBE Blend in II Ethanol* in
Total Mixture || Total Mixture || Mixture
0 .0
2 .5
5.0
7.5
10 .0
15.0
20 .0
25.0
30 .0
40 .0
50 .0
60 .0
80 .0
100 .0
100 .0
97.5
95.0
92 .5
90 .0
85.0
80 .0
75.0
70 .0
60 .0
50 .0
40 .0
20 .0
0 .0
0 .0
0 .25
0 .5
0 .75
1.0
1.5
2 .0
2 .5
3 .0
4 .0
5.0
6 .0
8 .0
10 .0
Total Percent of
MTBE in Mixture
11.0
10 .725
10 .45
10 .175
9.9
9.35
8 .8
8 .25
7.7
6 .6
5.5
4 .4
2 .2
0 .0
 * Denatured without methanol
       TABLE #4     GENERAL  MOTORS  RVP  TEST DATA

STUDY 1
STUDY 2
BASE RVP
9.4
12 .7
7.0
9.0
14 .0
ARVP
1.1
0 .9
1.1
1.0
0 .7
                                             44

-------
III.  RESULTS AND  CONCLUSIONS


      Figures #8  and  #9 show the results of the commingling model for a base
RVP of 8 psi.  Figure #8 shows the RVP boost due to the  commingling effect
that  can be expected  across all the gasoline in the fleet  based on ethanol
market shares of  10%, 30%, or 50%.  For each market share  a number of runs
were  performed to incorporate results from the various loyalty and fill curves
that  are available.   Figure #9 provides what is called the "Effective
Commingling Boost", which is defined as the fleetwide RVP  boost due to
commingling divided by the fraction of the fuel represented by the ethanol
blend.

      The results for an individual ethanol market  share appear to be most
dependent upon the type of loyalty curve chosen.  For each ethanol market
share the lowest  commingling effect is calculated using  the Arco brand loyalty
data  which is considered to be the high end of the  loyalty curves used.  The
highest commingling effect is found when no brand loyalty  is assumed.  The
other two brand loyalty curves provide similar results that are in the middle
of these two extremes.  It is believed that the these curves successful
bracket the "actual"  loyalty curve and thus these results  should bracket the
"actual" commingling  effect33.

      The results appear to be fairly insensitive to the type of fill curve
used.  In some instances there is little or no difference  in the commingling
effect between the use of the actual General Motors fuel refill data or the
smoothed version  of that data.  Thus this variable  appears to be not as
critical a factor as  does the owner loyalty to a fuel brand.

      The results do  show, however, that commingling can be a significant
effect.  The effect appears to be greater as the ethanol market share becomes
large  (Figure #8), however, the increase in the commingling effect appears to
level off at a maximum as the ethanol market share  becomes 30% or greater.
The effective commingling boost per gallon of ethanol blend, however, is the
greatest at low gasohol market shares.

      One of the  most dramatic points in these results is  the fact that the
non-reformulated  gasoline scenario results in a much greater commingling boost
than  that of the  reformulated gasoline case).   This is due to the fact that
the presence of MTBE  mitigates the RVP boost somewhat at low concentrations of
ethanol as is evident from Figure #6.
      33  The term  "actual"  in  this  case does not imply that there
is  one  loyalty curve that  is  applicable.   As was discussed
earlier  the loyalty  curve  should vary from one geographical
area to  the next depending on availability of  gasoline stations
and other variables.

                                     45

-------
Insert page of graphs here
                                      46

-------
Insert page of graphs here
                                      47

-------
Insert page of graphs here
                                      48

-------
II.  Simple Model


A.   Simple Model Equations and Referenced Work

1.   Baseline Determination

     The Act requires EPA to promulgate standards for the
performance of reformulated gasoline that are relative to
emission levels from baseline vehicles using baseline fuel.  In
order to determine whether fuels are meeting the performance
requirements of reformulated gasoline under the simple model, EPA
must, therefore,  establish the baseline to which fuels are to be
compared.  The following discussion describes how EPA derived the
emission baselines.

a.   Control Periods

     Before the emission baselines can be determined, the time
frame to which a fuel's performance will be compared must first
be identified.  Section 211(k) of the Act requires control of VOC
emissions during the "high ozone season."  For the purposes of
this rulemaking,  the high ozone season is defined to be June 1
through September 15.  This period covers the vast majority of
days during which the national ambient air quality standard for
ozone is exceeded nationwide and is consistent with the period
covered by EPA's gasoline volatility control requirements.  Fuel
will be required to meet the VOC control requirements at the
retail outlet from June 1 through September 15 and at the
refinery from May 1 through September 15.  Additional VOC
controlled fuel is likely to be produced by the refiner prior to
May 1 to ensure compliance on that date.

b.   Baseline Gasoline

     The Act specified the baseline fuel to be used in
determining compliance with the reformulated gasoline performance
requirements during the summer months (high ozone season).  That
specification is shown below.  The Act did not specify the
requirements for the winter baseline (non-high ozone season)
gasoline, but requires that EPA establish the winter baseline
specifications to reflect gasoline qualities in 1990.  The July
9, 1991  (1) proposal contained a description of the data and
methodology EPA used in determining the winter baseline fuel.
That methodology was revised somewhat by the April 16, 1992  (2)
proposal.  The final winter baseline fuel specification is shown
below, consistent with the non-high ozone period described above.
Beyond the requirements of the Act, and in keeping with EPA's
July 9, 1991 and April 16, 1992 proposals for reformulated
gasoline, the summer baseline fuel is also assumed to be free of


                                49

-------
oxygen and lead, and contain deposit control additives sufficient
to meet the requirements of Section 211(1)  of the Act.

Table II-l:  Baseline Fuel Compositions

                                   Summer    Winter
     Sulfur,  ppm                   339       338
     Benzene,  volume percent       1.53      1.64
     RVP,  psi                      8.7       11.5
     Octane,  R+M/2                 87.3      88.2
     T10,  degrees F                128       112
     T50,  degrees F                218       200
     T90,  degrees F                330       333
     Aromatics, volume percent     32.0      26.4
     Olefins,  volume percent       9.2       11.9
     Saturates, volume percent     58.8      61.7


c.   Baseline Vehicles

     Under Section 211(k)  of the Act,  reformulated gasoline must
result in reduced emissions relative to emissions from baseline
or representative model year 1990 vehicles when using baseline
gasoline.   For the purposes of this rulemaking, EPA has based its
emission modeling on vehicles having technology representative of
that used in 1990.  As a result, 1990 technology vehicles could
be as early as 1986 model year vehicles, or as late as current
model year vehicles.  The data used to develop the simple model
relationships being promulgated today are based on this.  For
more information regarding 1990 baseline vehicles, the reader is
referred to the Docket.34
2.    In-Use Basis

a.    MOBILE4.1 and Enhanced I/M

     The goal of EPA in developing the procedures for certifying
fuel as meeting the reformulated gasoline requirements is to
assure that a certified fuel will achieve the required emission
reductions in-use.  This goal necessitates the use of a fuel
effects model which predicts in-use emissions.  For the simple
model, EPA proposed to use the MOBILE4.1 emissions model to
determine the baseline emission levels for 1990 technology
vehicles over their full useful life.  Since the existence of an
enhanced inspection and maintenance  (I/M) program will
significantly effect the in-use emission performance of 1990
     34    Memorandum from Dave Korotney  to  Air Docket A-92-12
entitled "1990 Baseline Vehicles",  November 30,  1993.

                               50

-------
technology vehicles during the period when the reformulated
gasoline program is in place, EPA also proposed to assume an
enhanced I/M program was in place before determining baseline
emissions.  As discussed at length in the February 26,  1993
proposal, a more recent MOBILE model now exists,  and EPA has now
promulgated the requirements for enhanced I/M programs.  The
latest MOBILE model and actual I/M program requirements differ
significantly from those assumed for the proposal.  However,
given the impact of changes to these assumptions on the relative
benefits of exhaust and evaporative controls and the short lead
time available for refiners to respond to any such changes, the
simple model as promulgated here remains based upon the July 29,
1991 version of MOBILE4.1 and an enhanced I/M scenario consisting
of a gas cap check and a 2500 rpm idle test.  EPA believes this
is appropriate for the same reasons relied on to delay mandatory
use of the complex model until January 1, 1998.

b.   Temperature Conditions

     MOBILE4.1 was developed to predict motor vehicle emissions
on an area-specific basis.  In order to use MOBILE4.1,  it is
necessary to specify a temperature range for the areas in which
motor vehicle emissions are being evaluated.  For purposes of
reformulated gasoline, emissions are modelled during the summer
assuming temperatures ranging from 71.6 to 91.6 degrees
Fahrenheit in VOC control region 2 (northern areas) and ranging
from 69.4 to 94.0 degrees F in VOC control region 1 (southern
areas).  These temperatures represent the population-weighted
average of minimum and maximum temperatures measured in each of
25 serious and worse ozone nonattainment areas during their ten
worst ozone days in each of the months of July and August for the
years 1986 to 1989 (in ten of the cities) and 1985 to 1987  (in
the other fifteen cities) ,35  (3)

     Refueling emissions were derived assuming an ambient
temperature of 90°F for both VOC control regions 1 and 2.
Distinguishing between the different areas did not appear
justified given the similarity of Class B and Class C area
temperatures, the relatively low magnitude of refueling
emissions, and the wide range of times and temperatures at which
refueling occurs during a day.  90°F was considered to represent
a severe case in order to account for average in-use refueling
emissions on high ozone days.

     For determination of winter baseline emissions, an average
low temperature and an average high temperature of 39°F and 57°F,
respectively, are assumed.  These temperatures were estimated
     35 Memorandum  II-A-2  from Jeffrey A.  Herzog and Stephen
Mayotte to Public Docket  No.  A-91-02.

                                51

-------
from the historical 30-year average low and high temperatures for
the months of October through April for the 25 serious and worse
ozone nonattainment areas.36  (3)

c.   Effects of Stage II Refueling Controls

     The assumptions used in establishing the baseline emissions
should be representative of the conditions that will be
encountered during the time when the reformulated gasoline
provisions are in effect.  For this reason, baseline emissions
are assumed to include the benefits of a Stage II refueling vapor
recovery program.  By 1995, all serious, severe,  and extreme
ozone nonattainment areas are required to have implemented Stage
II refueling control programs.  The efficiency assumed for Stage
II controls in determining the emission baseline is 86 percent.
EPA's regulatory impact analysis supporting refueling emission
regulations estimated the efficiency of Stage II equipment to be
86 percent in areas such as California where the program is very
strictly enforced.  Because of the severity of ozone pollution in
areas that will be covered by the reformulated gasoline program
and because strong measures will be required to bring these areas
into attainment,  it is assumed that Stage II programs in these
covered areas will be strictly enforced.  For these same reasons
EPA is not revising its proposed efficiency assumption for Stage
II programs as suggested by some commenters.

3.   Emission Effects

a.   Definition of VOC

     The Act requires reductions in emissions of ozone-forming
VOCs.   This interpretation is consistent with the focus of
Section 211(k) on the areas with the most extreme ozone pollution
problem.  EPA proposed in the April 16, 1992 SNPRM that methane
would be excluded from the definition of VOC on the basis of its
low reactivity in keeping with past EPA actions,  but included all
other VOCs including ethane.  EPA further proposed, however, that
should the Agency modify the definition of VOC, we might do so
for the reformulated gasoline rulemaking as well.  As discussed
in the February 26, 1993 proposal, EPA has also excluded ethane
from the definition of VOC  (57 FR 3941).  As a result, the
performance of fuels meeting the VOC emission requirements under
the simple model, as expressed above,  are determined on a non-
methane, non-ethane basis.  This change resulted in slight
changes to the simple model equations previously proposed, but
the overall results of the simple model are essentially
unaffected.  The ethane emissions were subtracted from the SNPRM
estimates of the VOC emissions.  From the Auto\0il Air Quality
       Ibid.

                                52

-------
Improvement Research Program (AQIRP),  ethane emissions were
determined to be 3.5 percent of the nonmethane hydrocarbon
emissions.  Ethane emissions were assumed to be unaffected by
fuel changes.

b.   VOC Emission Effects

     At the time of the proposed rule the impact on VOC and NOx
emissions of adding oxygenates to gasoline was calculated using
the EPA emission factor database.  This database included both
normal and high emitting vehicles which are represented in the
in-use fleet.   The understanding at the time was summarized in an
EPA memorandum. (4)  The oxygenated fuels used in that analysis
were lower in aromatics level as is expected with the use of the
simple model.

     The nonexhaust VOC emission effects were determined using an
early version of the MOBILE4.1.  The only parameter that had an
effect on emissions was assumed to be the Reid Vapor Pressure
(RVP).  The MOBILE4.1 model was run for several RVP levels and
regressed to produced the equation given in the regulations for
VOC emissions.

c.   Exhaust Benzene Baseline Emissions and Effects

     The exhaust benzene emissions can be affected in two ways.
Reduction of exhaust VOC emissions through improved combustion or
catalyst efficiency should result in reduction of benzene
emissions.  Also,  the benzene fraction of VOC emissions may be
reduced either by reduction of fuel benzene or precursors
(primarily nonbenzene aromatics) of exhaust benzene.  This method
is used for the aldehyde emissions as well.

     The benzene fraction of exhaust VOC emissions was determined
from a correlation developed by Chevron using fuel benzene and
aromatics content as parameters. (5)   The Chevron analysis used a
compilation of all of the studies available at the time of the
proposal.

     Since exhaust VOC emissions are reduced with added
oxygenate, benzene emissions are assumed to be reduced
proportionally.

     Low temperature testing indicated that the benzene and 1,3-
butadiene emissions fraction were found to be insensitive to
temperature.   (6,7)  Therefore, the winter baseline emissions
should be expected to increase as the exhaust VOC emissions
increase.
d.   Nonexhaust Benzene Emissions
                                53

-------
     Benzene is the only toxic air pollutant that is emitted in
measurable quantities from refueling, evaporative, running loss
vapors.  Reduction in fuel benzene can be expected to result in
proportional reductions in benzene emissions from all of these
nonexhaust emissions sources.

     The nonexhaust benzene emissions were determined by applying
the GM headspace benzene fraction estimate to the mass emissions
estimated from the MOBILE model.(8)   The MOBILE model was used to
predict the RVP effect on VOC emissions down to as low as 6.6
psi.   Due to differences in the temperature conditions,  slight
differences in nonexhaust VOC emissions occur between VOC control
region 1 and 2.   The GM headspace model showed an effect of both
RVP and MTBE content on the fraction of VOC emissions that were
benzene.

     Evaporative benzene emissions include hot soak  (evaporative
emissions immediately following the shutting off the engine) and
diurnal emissions  (evaporative emissions from a sitting vehicle
as the daily ambient temperature rises).   Hot soak emissions
occur at higher temperatures than diurnal emissions and the
relative volatility of benzene is slightly higher at those
temperatures.  Running loss emissions occur at roughly the same
fuel  temperatures as hot-soak emissions and have similar benzene
fractions.

     Based on the emission factors contained in MOBILE4.1,
evaporative and running loss benzene emissions tend to be
dominated by emissions from vehicles with inoperative emission
control systems (those vehicles likely to "fail" EPA's purge and
pressure tests).   The benzene fraction of evaporative and running
loss  emissions from vehicles with properly operating systems
(those vehicles likely to "pass" EPA's purge and pressure tests)
and from "fail"  vehicles, however, are comparable.  Hence the
benzene weight fraction of evaporative and running loss VOC
emissions for a fuel is assumed in the simple model to be
described by the relationship originally derived for "fail"
vehicles.

     One comment suggested that it was not appropriate to use
estimates not publicly available,  like the GM model for
estimating emission impacts.  While the GM model itself is not
available,  the emission estimates from the model are available.
Furthermore, the underlying basis of the model has been published
by Reddy (9), and a separate program developed by EPA using
Reddy's paper as the basis confirmed the estimates from the GM
model.   Another comment suggested that the headspace model
underestimated the benzene emissions with MTBE present.   It was
explained that the higher average molecular weight of the
headspace vapor partially offsets the benzene fraction reduction.
If the vapor has a higher molecular weight,  it could be concluded


                                54

-------
that the VOC emissions would be greater.  However insufficient
data existed to quantify the effect of MTBE on evaporative
emissions at constant RVP.   EPA believes it is not appropriate to
alter the benzene fraction estimate,  since there is no
quantitative estimate for the mass emission increases claimed by
the commenter.

     Under winter condition, EPA is assuming that nonexhaust
benzene  (and all nonexhaust VOC)  emissions will be negligible
relative to exhaust toxic emissions.   A comment suggested that
the MOBILE model will predict emissions for wintertime conditions
at the higher RVP levels used,  and therefore should be included
in the toxic model.  While the RVP of the fuel is much greater
during the winter, the temperatures are much lower.  EPA's MOBILE
model is based upon data collected above the temperatures
experienced in the winter and generally with RVP levels below
those of winter fuels.  Projecting evaporative emissions in the
winter is outside of the capability of the MOBILE model.  Biven
the lack of adequate data to model nonexhaust VOC and benzene
emissions in the winter, EPA is not changing from its proposed
baseline and toxics model in this regard.

e.   Aldehyde Baseline Emissions and Effects

     The baseline emissions for the aldehyde emissions,
formaldehyde and acetaldehyde,  were determined from the AQIRP
results.  The baseline emission estimates are slightly different
than the SNPRM values to account for the change in the definition
of VOC emissions from the SNPRM which excluded ethane.  While a
comment expressed a concern that AQIRP results should not be used
exclusively, only aldehyde emissions effects were determined
using the AQIRP data and those exclusively because no other data
existed at the time.

     The results from AQIRP were used to determine the effects on
the formaldehyde and acetaldehyde fraction of exhaust VOC from
the addition of oxygenates.  Explicit effects were calculated for
MTBE, ethanol, and ETBE.

     The AQIRP data as released,  however, were modified slightly
to exclude the acetaldehyde and formaldehyde results for ETBE and
ethanol from one of the vehicles (car #5A) due to emission
results which were confirmed as being in error.  Furthermore, the
effect of ETBE on the weight percent of acetaldehyde was based on
the test results for ethanol due to the lack of adequate fuel
comparability for ETBE-containing fuels.  However, for this case,
the ethanol results were adjusted based on a comparison of the
ETBE and ethanol results on similar fuels for which data was
available.

     Responding to several comments,  the EPA has included
aldehyde equations for the use of other oxygenates in the simple

                               55

-------
model.  Methyl tertiary-amyl ether (TAME)  and any other methyl
ethers are assumed to have equivalent aldehyde emission
performance to MTBE.   Ethyl tertiary-amyl ether (ETAE)  and any
other ethyl ethers are assumed to have equivalent aldehyde
emission performance to ETBE.   Alcohols (with the exception of
methanol)  are assumed to have equivalent aldehyde emission
performance to ethanol.   These assumptions were based on the
chemical similarity of the oxygenates.  Aldehyde equations for
mixed oxygenates were also provided.

f.   Butadiene Emissions and Effects

     At the time of the proposal there was little information on
the effect of fuel reformulation on butadiene emissions.  A
reduction in exhaust VOC emissions was assumed to result in a
proportional reduction in butadiene emissions.  The baseline
emissions were determined from an analysis of the AQIRP.

     For wintertime modeling,  the baseline emissions were assumed
to be the same percentage of the VOC emissions as the summertime
emissions.  See the benzene discussion above for the rationale.

g.   POM Emissions and Effects

     There was no data to determine the effect on POM emissions
from the use of reformulated gasolines.  An assumption was made
that the reduction of POM emissions would be the same as the
reduction of VOC emissions.

     The baseline emissions of POM were determined from work of
Schuetzle (10).   Table 8 in that paper shows that 18 mg/km of
particulate were produced from gasoline vehicles (half of the
vehicles were equipped with catalysts, the HC emissions were not
reported).  Approximately 8.5 mg/km of the particulate was
extractable in three separate extracts, (1% ether in hexane
(nonpolar fraction),  1/1 ether/hexane  (moderately polar),  and
methanol (polar fraction).   Selected standard PAH compounds were
recovered at greater than 80%.

     The paper indicated the approximate percentage of PAH which
is interpreted as synonymous with POM in each fraction for diesel
exhaust.  The extract from gasoline exhaust was found to be more
mutagenic on a gram basis than diesel extract therefore we may be
underestimating the POM emissions by applying the same percentage
of POM in diesel particulate extract to that for gasoline
particulates.

     Gas phase POM were assumed zero because the mutagenicity was
low.  The polar fraction was assumed zero because activity was
low and no data was presented.  Both of these assumptions lead to
a lower estimate of the POM emissions than might have been.


                                56

-------
     The sum of the nonpolar and moderately polar extract
fraction multiplied by the POM fraction in each extract provides
an estimate of the fraction of POM in the particulate phase.
This estimate multiplied by the amount of particulate extract for
gasoline engines provides the 1.4 mg/mi baseline emission
estimate of POM.

     One comment cited two references which estimated lower POM
emissions.  Schuermann, et al.  (1990) (11) reported that 0.2
mg/mi and 0.01 mg/mi of PAH were emitted from cars without and
with a catalyst respectively.  Only 11 PAH compounds were
identified and quantified.  Whether these compound were collected
from the gas phase or the particulate phase was not discussed in
the paper and a more extensive literature review is necessary.
The HC emissions were not reported.  Westerholm, et al.  (1988)
(12) reported that gas phase PAH emissions from a car without a
catalyst were 0.4 mg/mi, and particulate associated PAH was 0.07
mg/mi.  The HC emissions were 2.4 g/mi,  and the particulate
collected was 7.3 mg/mi.  The particulate was extracted with
dichloromethane (methylene chloride)  and no recovery of selected
PAH compounds were reported.  This raises questions regarding the
choice of solvent especially given the low value of recovered PAH
from the amount of particulate.


B.    Technical Corrections

     Several errors were found in the proposed simple model.
Baseline values were found to be in error for POM emissions and
Region 1 total VOC and toxic emissions.   The baseline emissions
for the final simple model are given below.  Any slight
difference between the final baseline levels and those in the
proposal are due to the round-off differences when ethane was
excluded from the VOC emissions.
                                57

-------
Table II-2:   Simple Model Baseline Emissions

                              Summer
Exhaust VOCs (g/mi)
Hot Soak/Diurnal (g/mi;
Running Loss (g/mi)
Refueling (g/mi)
Total VOCs  (g/mi)

Exhaust Benzene  (mg/mi;
Evaporative Benzene
Running Loss Benzene
Refueling Benzene
1,3-Butadiene
Formaldehyde
Acetaldehyde
POMs
Total TAPs  (mg/mi)
                         Class B
                         0.444
                         0.390
                         0.431
                         0.040
                         1.304

                         30.1
                         4.3
                         4.9
                         0.4
                         2.5
                         5.6
                         4.0
                          1.4
                         53.2
Class C
0.444
0.339
0.390
0.040
1.213

30.1
 3.8
 4.5
 0.4
 2.5
 5.6
 4.0
 1.4
52.1
                                             Winter
0.656
0.656

40.9
 0.0
 0.0
 0.0
 3.6
 5.6
 4.0
 1.4
55.5
C.
Caps on Fuel Parameters
     The caps on three parameters  (sulfur, T90, and olefins) were
chosen based on the information at the time indicating that
increasing these parameters increased either VOC or NOx
emissions.  Subsequent comments felt that aromatics and T50
should be capped in addition to the three proposed.  Other
comments felt that no properties be capped.  In addition, some
comments felt that the driveability index  (a combination of T10,
T50, and T90 with the most emphasis on T50 and T90) should be
included in the simple model or at least capped.

     The basis for the caps was information from the AQIRP
technical bulletins (13, 14) which showed that T90 had a
significant impact on VOC emissions, olefins on NOx emissions,
and sulfur on both.  Since allowing these parameters to increase
over the baseline value was certain to have a detrimental effect
emissions, it was necessary to cap them.  Since that time
additional data and information has been developed for use in the
Complex Model.  This data confirms that sulfur, T90 (E300) and
olefins should be capped either for the purposes of preventing
increases in VOC or NOx.

     Aromatics and T50  (E200)  have also been found to have a
significant effect on emissions.  As a result, they may also be
deserving of caps.  However, in the case of aromatics, a separate
cap is less important since its level is limited by the toxics
requirements in the model.  In addition, for both aromatics and
T50, applying caps at this point in time would not be appropriate
                                58

-------
due to the need to provide refiners with adequate leadtime.
Similarly, regardless of the relative merits of using the
driveability index as a surrogate for or supplement to T10, T50,
and T90,  it would not be appropriate in the context of leadtime
constraints to place a cap on the driveability index of the fuel.


D.    Averaging and Trading

     Since the air toxics standard is a year-round standard,  as
long as unequal percent reductions are achieved in the summer and
the winter, the overall stringency of the standard is dependent
on the relative duration of the two periods.  In the proposals,
the time period over which refiners could count fuel as being
controlled for air toxics relative to the summer baseline was
sometimes internally inconsistent, and inconsistent between
proposals for various reasons.  As a result, EPA received a
number of comments on what the appropriate period should be.
Based upon the comments, the averaging period for the toxic
emissions performance standard will be the same as for the VOC-
controlled season.  This should avoid confusion and reduce
reporting time.

     Since the simple model does not include a NOx model, it
would not be possible to average NOx emissions as suggested by
one commentor.
                                59

-------
References

(1)   NPRM, 56 FR 31176, July 9, 1991.

(2)   SNPRM, 57 FR 13416, April, 16, 1992.

(3)   Memorandum II-A-2 from Jeffrey A. Herzog and Stephen Mayotte
          to Public Docket No. A-91-02.

(4)   Lindhjem, C.E., 'Effect of Oxygenates on Emissions', EPA
     Memorandum to Richard Rykowski,  January, 7, 1992.

(5)   Chevron Analysis

(6)   Stump, F., Tejada, S., Ray W., Dropkin, D., Black, F., Snow,
               R., Crews, W., Siudak, P., Davis, C.O., and Baker,
               L.,  'The Influence of Ambient Temperature on
               Tailpipe Emissions from 1984 to 1987 Model Year
               Light-Duty Gasoline Motor Vehicles-II',
               Atmospheric Environment, v. 23, 1989,  pp. 307-320.

(7)   Stump, F., Tejada, S., Ray W., Dropkin, D., Black, F., Snow,
          R., Crews, W., Siudak, P.,  Davis, C.O., and Carter, P.,
           'The Influence of Ambient Temperature on Tailpipe
          Emissions from 1985 to 1987 Model Year Light-Duty
          Gasoline Motor Vehicles-II', Atmospheric Environment,
          v. 24A, 1990, pp. 2105-2112.

(8)   Reddy, S.R., Personal Communication to C.E. Lindhjem and P.
          Bush, April 16, 1991.

(9)   Reddy, S.R., 'Evaporative Emissions from Gasolines and
     Alcohol-Containing Gasolines with Closely Matched
     Volatilities',  SAE-861556.

(10)  Schuetzle, D.,  (1983),  'Sampling of Vehicle Emissions for
     Chemical Analysis Biological Testing', Environmental Health
     Perspectives, v. 47, pp 65-80.

(11)  Schuermann,  D., et al.,   (1990),  SAE-902116.

(12)  Westerholm,  R.N.,  et al.,  (1988), Environmental  Science and
          Technology, 22, pp 925-930.

(13)  Auto\0il Air Quality Improvement Research Program, Technical
          Bulletin No.  1,  'Initial Mass Exhaust Emissions Results
          from Reformulated Gasolines', December, 1990.

(14)  Auto\0il Air Quality Improvement Research Program, Technical
          Bulletin No.  2,  'Effects of Fuel Sulfur Levels on Mass
          Exhaust Emissions', February, 1991.
                                60

-------
III. Complex Model Baselines

     Section 211(k)  of  the Clean Air Act  (CAA or the Act) requires
that gasoline sold  in  the worst ozone  nonattainment areas of the
United States be reformulated to result in reduced emissions of VOC
and toxic compounds.  The Act also states that the emissions from
gasoline in other  areas of the United States cannot simultaneously
increase above 1990 levels.  Refiners must certify their gasolines
for sale by showing, through testing and/or the Complex Model, that
their  fuel meets  the  requirements  of  either  the  reformulated
gasoline or  anti-dumping programs.   In  either  case,  the vehicle
emissions  must  be  evaluated  relative  to the  emissions  from  a
specified baseline gasoline  (outlined in Section  211(k) of the Act)
in a specified baseline vehicle  (defined by Section  211(k)  as  a
representative 1990 model year vehicle).   Thus the Act implicitly
directs  EPA  to  establish and  promulgate the baseline  emissions
against which all  emission measurements will be  evaluated.

     The baselines discussed in this section do not apply to the
cost-effectiveness  calculations presented in Sections  V  and VI.
Cost-effectiveness  is  based  on  in-use  emission  estimates,  and
therefore  the baseline  fuels  and baseline  emissions  cannot  be
limited to 1990 baseline fuels and 1990 model year vehicles  (i.e.
the  conditions  under  which  the  Complex  Model was  developed).
Alternative  baselines   will  be  derived  and presented in  that
section.

     Baseline  emissions are  produced  from  baseline  fuels  as
specified in the Act.  Accordingly,  this section  is composed of two
subsections,  one  for  baseline  fuels  and  a  second  for baseline
emissions.   This  section is directed  towards use of  the Complex
Model.    The baselines  applied to the Simple  Model  are covered in
more detail  in Section  II,  and baselines  applied  to  the anti-
dumping program are discussed in Section VII.


A.   Baseline Fuels

     The Clean Air Act  directs the EPA  to use a different baseline
fuel during different  seasons.   The Act provides the EPA with  a
specific baseline  fuel  for  the high ozone  (i.e.  summer)  season,
which EPA has defined  as the period  from June 1 to  September 15.
Although the Act  refers  to  this baseline fuel as  summer baseline
gasoline, its importance in the reformulated gasoline program has
resulted  in  its  being  commonly  referred to  as  Clean Air Act
Baseline Gasoline, or  CAAB.  The Act does not specify a baseline
fuel for the non-high ozone  (i.e.  winter)  season,  but  instead
directs the EPA to establish the parameters of such a baseline fuel
itself.  Both of these baseline fuels will be discussed below.
                                61

-------
1.
Summer
     Most of  the summer  baseline  fuel parameters  are specified
explicitly by the Act.  The level of oxygen, however, was left to
the EPA to determine.  As discussed in the July 1991 NPRM [56 FR
31176],   the  low  volume  of  oxygenated  fuels  sold  in  ozone
nonattainment areas  in  1990  led the Agency  to  conclude that the
oxygen content of the summer baseline fuel  should  be  zero.   The
final summer baseline fuel specifications are given in Table III-l.

               Table III-l:   Summer Baseline Fuel
Fuel parameter, units
Oxygen, wt%
Sulfur, ppm
RVP, psi
T50, °F
T90, °F
Aromatics, vol%
Olefins, vol%
Benzene, vol%
Value
0.0
339
8.7
218
330
32.0
9.2
1.53
In addition to the fuel  parameters  shown Table  III-l, the Act also
specifies  values  for  a  number of  other  summer baseline  fuel
parameters that are not included in the Simple or Complex Models:

               API gravity         57.4
               Octane,  R+M/2       87.3
               IBP, °F              91
               T10, °F              128
               End point,  °F       415
               Saturates,  vol%     58.8

     The distillation parameters shown in  Table III-l are given in
terms of T values, or the temperature at which a given percentage
of fuel has evaporated.   Distillation characteristics of baseline
fuel  (or  any  fuel)  can alternatively be  described  with E values
which represent the volume percent of fuel that has evaporated at
a given temperature.  Since the Complex Model uses  E values instead
of T values as distillation parameters, the Agency  developed a pair
of  conversion  equations.    Described in  more  detail  in Section
IV.A.2 below,  these equations are derived  from  a simple regression
on data from the Complex Model database.

                E200(%) = 147.91 - 0.49 x T50 (°F)
                E300(%) = 155.47 - 0.22 x T90 (°F)

where  E200 and  E300  are  the %  evaporated at  200 and  300 °F,
respectively,  and T50  and  T90 are the  temperatures at which 50 and
                                62

-------
90 percent of  the fuel has been evaporated,  respectively.  Thus the
summer baseline fuel distillation parameters  used by the Complex
Model are:

               E200(%)  =  147.91  - 0.49  x  218 =  41  %
               E300(%)  =  155.47  - 0.22  x  330 =  83  %
2 .
Winter
     The winter baseline  fuel was not defined  explicitly by the
Act.  Instead,  the Agency  was  directed to establish specifications
for winter baseline fuel.   As described in the July 1991 NPRM [56
FR 31176] and revised in  the  April  1992  SNPRM [57 FR 13416],  the
Agency based its winter baseline fuel specifications on an analysis
of fuel surveys performed  by the American Automobile Manufacturers
Association and by Southwest Research Institute.  The collection of
winter season values for each fuel parameter was weighted according
to sales data,  state  fuel  consumption data, and population data to
obtain  the  final, nationwide winter  baseline  fuel.   Thus  the
methodology and results  of this analysis remain unchanged from the
proposals.  The winter baseline  fuel specifications  are given in
Table III-2.

               Table  III-2:   Winter  Baseline  Fuel
Fuel parameter, units
Oxygen, wt%
Sulfur, ppm
RVP, psi
T50, °F
T90, °F
Aromatics, vol%
Olefins, vol%
Benzene, vol%
Value
0.0
338
11.5
200
333
26.4
11.9
1.64
     As for summer baseline gasoline, the distillation parameters
shown in Table III-2 are  given  in  terms  of  T values instead of E
values.  The conversion equations  described above  can be used to
estimate E  values for winter baseline  gasoline  as well  as  for
summer baseline gasoline:

               E200(%)  = 147.91  - 0.49  x  200 =  50 %
               E300(%)  = 155.47  - 0.22  x  333 =  83 %

3.    Response to Comments

     A number of commenters took issue with the Agency's decision
to define baseline fuel values for  E200 and  E300 through equations
                                63

-------
that provide  these  values  from T50 and T90  measurements.   These
conversion equations were based on  single parameter regressions of
all  the  data  in the  Complex  Model  database.   The  commenters
suggested that the  summer baseline  fuel values for  E200 and E300
should be  based instead on the measured  values for  these fuel
parameters  on  summer  baseline  gasoline  in  the  Complex  Model
database.  The average values as provided by the American Petroleum
Institute  are 41.9% for  E200  and  82.1%   for  E300.    The  Agency
recognizes the merit in  this approach.   However, the  Agency has
determined  that  the  use  of  the  conversion  equations for  the
baseline fuel  is  preferable because this  approach  is  consistent
with the use  of the same  conversion equations  for  other  fuels.
"Other fuels"  include the winter baseline  gasoline  (which  is not
represented in the Complex  Model database), fuels that are used in
the process of specifying performance standards, and fuels that are
candidates for certification with the Complex Model.  In addition,
while the "summer baseline" gasolines in  the Complex Model database
were blended so as to match the baseline fuel specifications laid
out  in the Act,  the  inevitable  variability  introduced  in  the
blending and measurement processes  (as discussed in Section IV.G)
caused the actual composition and properties of "summer baseline"
gasolines  in  the  Complex Model database  to vary from  the Act's
specifications.

     Commenters also suggested that the conversion  equations be
based  not  on  all  the  data  in the  Complex  Model database,  but
instead on the hydrocarbon-only fuels. The  Agency determined that:

     1)   It  was  not  appropriate  to  have  a  different   set  of
          conversion equations for hydrocarbon-only   fuels  and
          oxygenated fuels; and

     2)   It  was  not appropriate  to apply  conversion equations
          based   on  hydrocarbon-only   fuels   to   reformulated
          gasolines, all of which will be oxygenated.

Thus the Agency has  concluded that  its conversion equations, based
on all the  data  in the Complex Model database,  provide the most
reliable  means  of  estimating E200  and  E300  values  for  the
hydrocarbon-only  baseline   gasolines,   oxygenated   reformulated
gasolines,  and both  oxygenated and hydrocarbon-only conventional
gasolines.

     The Agency received many  comments objecting to  the use of a
7.8 psi RVP baseline fuel  for summer Class  B  areas.   The  Agency
agrees with the commenters that the Act  makes no provisions for a
summer baseline fuel with an RVP of 7.8 psi  and that such a fuel is
not an appropriate baseline  for determining performance standards.
Therefore,   the Agency  has  calculated all performance standards
relative to CAA baseline  fuel with an RVP of  8.7.   However,  the
Agency retains its position that a  fuel identical to Clean Air Act
baseline gasoline but  with an  RVP  of  7.8  psi provides  the most

                                64

-------
accurate measure of  in-use  emission  reductions  in Class B areas.
In such areas, EPA's volatility  control  requirements have capped
RVP levels at 7.8 psi.  Thus a gasoline with an RVP of 7.8 psi that
is otherwise identical to CAA baseline gasoline has been used for
the Class B cost-effectiveness calculations presented in Sections
V.B and VI.F.
B.   Baseline Emissions

     While baseline fuels are  categorized only by season, baseline
emissions are categorized by season, area class, and program phase.
They can be subdivided into exhaust and non-exhaust emissions and
by  pollutant.    Thus in  the  discussions  that follow,  baseline
emission derivations  and presentations will be given  within the
context of these category definitions.

     Baseline  emissions  are  based upon a  number of  derivation
approaches and sources.   The derivation approaches are specific to
the subcategories  described  above.   Of the sources,  the  MOBILE
emissions  model  is  the  most  important,  and  therefore will  be
discussed first in  Section B.I.  The following two sections outline
the scenarios under which the  MOBILE model was run.  The weighting
factors used to  calculate toxic  emissions will  then  be discussed
and presented in Section  B.4.  The  last two subsections of Section
B  will  present  the baseline  emissions for  summer  and  winter,
respectively.

1.   MOBILE Emissions Model

     The Agency's MOBILE  emission factor model  is an integrated set
of FORTRAN routines  for  use in  the analysis of  the air pollution
impact  of  gasoline  and   diesel  fueled  highway  mobile  sources.
MOBILE calculates  emissions in  terms  of grams per mile  for both
gasoline and diesel-fueled light-duty vehicles (LDVs),  light-duty
trucks  (LDTs), and  heavy-duty vehicles (HDVs),  and for gasoline-
fueled motorcycles.  These emissions,  or emission factors,  can be
calculated in  any  one of  five  forms:  total  hydrocarbons   (THC),
nonmethane hydrocarbons  (NMHC),  volatile organic compounds  (VOC),
total organic  gases  (TOG),  or  nonmethane  organic gases  (NMOG).
These categories are defined in  Table III-3.    The  effects  of
various inspection and maintenance programs  can also  be modeled.
                                65

-------
         Table III-3: MOBILE Emission  Factor Definitions
Emission
Factor
Option

THC
NMHC
VOC
TOG
NMOG
Compounds included in the emission factor:

FID
Hydrocarbons
Yes
Yes
Yes
Yes
Yes


Methane
Yes
No
No
Yes
No


Ethane
Yes
Yes
No
Yes
Yes


Aldehydes
No
No
Yes
Yes
Yes
     When quantifying  emissions from  a  fleet of vehicles,  it  is
important to  realize that  emissions will  vary over a  vehicle's
lifetime, generally  increasing with age due  to  factors  including
normal mechanical aging and possible malmaintenance or  tampering.
To ensure that the standards set for reformulated gasoline achieve
the intended environmental benefits, the Agency assesses emissions
on an  "in-use" basis,   such  that emissions  from 1990 model  year
vehicles are  the  estimated average emissions  from  those vehicles
over their lifetimes.

     MOBILE  assumes  that  a  25-model-year  "window"  of  vehicles
comprises  the  in-use  fleet  in any given  calendar year.    Any
vehicles over  25 years of age  are grouped  with and  modeled as the
25-year-old  vehicles.    For  example,   if   the  calendar year  of
evaluation were 1990,  then  MOBILE would  assume that the  in-use
fleet would consist of vehicles ranging in  age  from one (model year
1990) to 25 (model year 1966 and older).

     EPA's MOBILE emissions model  is designed to predict  motor
vehicle emissions on an area-specific  basis.   In order  to use the
MOBILE model,  it  is  necessary to specify  a  temperature  range for
that area in  which motor vehicle  emissions are  being  evaluated.
The area distinctions assumed by EPA are  roughly consistent with
the  ASTM area classifications  A,   B,  C,   D,  and  E.   For  this
analysis,  a  diurnal  temperature  range  of  72  to  92  degrees
Fahrenheit was used  for Class  C  areas,  and  a range of 69  to  94
degrees Fahrenheit was used for Class B areas.  These temperatures
represent the  average minimum  and maximum  summer temperatures for
high ozone  days  in  these  areas.    For determination   of  winter
baseline emissions in  the non-high  ozone  period, an average low
temperature  and an  average  high  temperature of  39°F   and  57 °F,
respectively,  were assumed for both  B  and  C areas.37
     37
       EPA Memorandum from Jeffrey Herzog and Stephen Mayotte to Public
Docket No. A-91-02,  "Summer and Winter Temperatures for the Top 25 Serious,
Severe,  and Extreme  Ozone Nonattainment Areas"
                                66

-------
     MOBILE  calculates  nonexhaust  emissions  in three  separate
categories:  running losses,  refueling  losses,  and  evaporative
losses.  Running losses  are  losses that occur while the vehicle is
running, and are brought about  by the heating of  the fuel tank and
fuel lines.  Refueling losses occur strictly during refueling of a
vehicle, and are due mainly to  the displacement  of  vapors  in the
fuel tank with fuel.  Evaporative losses include both diurnal and
hot soak losses.  Hot soak losses occur  immediately after a vehicle
is turned off,  and are due to the continued transfer of heat from
the engine to the fuel and the concurrent removal of cooling media
such as wind and  engine coolant.  Diurnal  losses occur during daily
ambient temperature swings.

     In the February 1993 NPRM  [58  FR 17175],  MOBILES was used to
determine baseline  emissions of VOC  and NOx for Phase  II  of the
reformulated gasoline  program.   Since  then,  MOBILE5a has  been
released,  and  it provides  slightly  different baseline  emission
estimates  than  MOBILES.  The   changes  to  the baseline  emission
estimates are minimal,  however, prompting the Agency  to  replace
MOBILES with MOBILESa without opportunity for public comment.

     MOBILESa was initially designed to augment MOBILE4.1 through
the addition of  post-1990 vehicle based  emission controls.   Both
MOBILE 4.1 and MOBILESa were intended  to predict the same  in-use
emission for calendar years prior to 1991.  However, in the process
of adding  the post-1990 vehicle controls,  EPA also  modified its
estimates of pre-1991 model year vehicle emissions, causing the two
models to  produce  differing projections for pre-1991  as well as
post-1990 calendar year  emissions.  The  most significant changes in
MOBILESa from MOBILE4.1  include:

     •    The effect of post-1990 vehicle  and  fuel  emission
          control programs mandated by  the  Clean  Air Act have
          been added;
     •    The California low-emitting vehicle program can be
          modeled;
     •    A  wider   range   of   inspection   and  maintenance
          programs  can  be  modeled,  most  notably  programs
          meeting EPA's  definition of "enhanced  I/M."
     •    The basic emission  rate  equations  for 1981  and
          later  light-duty  gasoline  vehicles   have   been
          revised,  based  on results  of  the  IM240  transient
          test program at Hammond,  IN.
     •    Speed  correction  factors  for  light-duty  gasoline
          vehicles have  been revised.

Since the reformulated gasoline program uses 1990 vehicles for its
baseline, only those changes that affect 1990 model  year vehicles
will affect the MOBILESa output and therefore  the  Phase II baseline
emissions.  Thus, only the last  two points  above  contribute to the
difference in baseline values between MOBILE4.1 and MOBILESa.  The
significance of  these changes in the MOBILE  model  on the derivation

                                67

-------
of the Complex  Model  for nonexhaust emissions  is  demonstrated by
the increase in total emissions from MOBILE5a over MOBILE4.1 of 13
percent,38 when  a  basic I&M scenario is assumed for MOBILE4.1  and
an enhanced I&M scenario is assumed  for MOBILE5a.

      For  the  simple model, the Agency derived  baseline  emissions
using a July 11,  1991  version  of EPA's MOBILE4.1  emissions model.
As described  below, EPA also  used  this  version  of MOBILE4.1  to
generate  the  baseline  emissions   for  all  of  Phase  I  of  the
reformulated gasoline program (i.e. through 1999).  Thereafter,  the
more recent MOBILE5a model will be used.

2.    Inspection and Maintenance Programs

     The  baseline  emissions  used  in   the   simple   model were
determined from MOBILE4.1  under a scenario that included an early
definition of  enhanced  inspection  and maintenance  (I/M).  This
early version of  enhanced  I/M  consisted of a short idle  test  for
exhaust emissions and  a visual check for  tampering or  malfunction
of the catalyst, the fuel inlet restrictor, and  the gas cap.  Since
publication of  the  simple  model SNPRM, the Agency has  promulgated
regulations for enhanced I/M programs that diverge in structure  and
content from  the  original  definition of  enhanced I/M.   The more
recent enhanced  I/M  program  includes a  dynamometer-based  IM240
exhaust emissions test and fuel  tank pressure  and  canister purge
checks  for  the  vehicle's nonexhaust  emission control  systems.
Under the model enhanced I/M  scenario,  98 percent of  the vehicle
population will undergo  inspections  once  a year,  and waivers will
be granted for  1  percent of  the  inspected vehicles.   The testing
procedures for the emission factor evaporative system pressure  and
purge checks,  as  well as  that for  the  IM240 transient  test,  are
provided  in  the  final rulemaking on  inspection  and  maintenance
program requirements   [57  FR  52950].   The  enhanced   I/M  program
assumed in the April 1992  SNPRM for  the simple model [57 FR 13416]
would now be considered more representative of  "basic" I/M.

     The  Clean Air Act  requires  ozone  nonattainment  areas  to
achieve the primary standard (less than 0.121 ppm ozone)  within a
specified time  period  depending  on the area classification.   The
Agency's final rule on inspection and maintenance  programs directs
all states or areas within an  ozone transport  region  that have a
1990  population  of  100,000 or  more  to  implement  enhanced  I/M
programs  to  achieve   that  standard,  regardless  of  that  area's
attainment classification.   Serious or worse ozone nonattainment
areas having a 1980  population  of 200,000 or more also are required
to implement  enhanced I/M programs.  Although all of these enhanced
         Based on total (exhaust + non-exhaust)  non-methane, non-ethane VOC
emissions  for light-duty cars (not including resting losses) in Class C areas
using an 8.7 RVP fuel.  MOBILE model input files  are shown in Appendices III-A
and III-B.

                                68

-------
I/M programs  are required to  begin by January  1,  1995, various
aspects of the program can be phased in over time.  Thus, EPA does
not  anticipate  the  full  impact  of  enhanced  I/M  to  be  fully
effective until  January  1, 1999,  which is late in Phase I of the
reformulated gasoline program.  Given that the complex model can be
used as early as  1995, the simple model's basic I/M program appears
to represent  Phase  I of the  reformulated gasoline  program best,
while the recently promulgated enhanced I/M is more representative
of I/M programs  in Phase II.

     Under the basic I/M program  scenario,  MOBILE4.1 predicts that
20 percent of the gasoline-fueled light-duty  vehicles on the road
would  be   "failing"  vehicles39   from  a  nonexhaust  emissions
standpoint.  A nonexhaust failing vehicle  is  defined as one whose
evaporative  emission control  systems  do  not work  properly,  as
determined by the pressure  and  purge  checks  associated with an
enhanced I/M program.  This scenario is the one assumed for Phase
I.  Under the auspices of an  enhanced I/M  program  during  Phase II,
MOBILESa predicts that essentially every light-duty vehicle on the
road would be designated  as  "passing"  in  terms  of evaporative
emissions  because  enhanced I/M  programs  are assumed  to have 98
percent  compliance  and  to grant waivers  for only  1  percent of
failing vehicles.

3.   Stage II Refueling Controls

     By 1995, all serious and worse ozone  nonattainment areas will
be required to implement Stage II refueling controls  as a means of
limiting gasoline refueling emissions.  These controls are designed
to capture displaced vapors from the vehicle  fuel tank and return
them to the underground fuel  storage tanks at  the  service station.
In the  Draft  Regulatory  Impact Analysis for the simple model, Stage
II  refueling controls  were   estimated to  decrease  uncontrolled
refueling  emissions  by  73  percent.  This effectiveness  has been
carried over to  the complex models  for both Phases I and II.

     On-board refueling  controls will  begin  in  1996.   Since the
Complex Model is  concerned only with 1990-technology vehicles, on-
board refueling  controls have no bearing  on baseline emissions or
on emission estimates using the Complex Model.

4.   Weighting Factors

     The  statistical regressions that  were used to  develop the
Complex  Model produced separate  models  (equations)  for  every
vehicle  group   for   each  of  six  exhaust  pollutants.    This
categorization of  emission effects  was done  to  most accurately
describe  the  unique  combustion   chemistry  characteristics  of
     39
        Communication with Lois Platte,  EPA National Vehicle and Fuel
Emissions Laboratory, Sept. 21, 1992


                                69

-------
different types of  vehicles.   However,  these  separate equations
must be  weighted according  to the  in-use  contribution  of  each
vehicle  type  to  the total  mobile  source emissions  inventory in
order to determine the  overall  in-use performance of a given fuel.
The derivation of  these  weighting  factors will be  given  in  this
subsection.    The  normal  emitter  in-use contribution  of  each
technology type  to the  total mobile source VOC emissions inventory
is also used in deriving  toxic  emission baselines, as described in
Subsection III.B.5.

     EPA subdivided normal emitting vehicles  (those that emit less
than 0.82 g/mi of VOC) into nine broad vehicle technology types.
Each technology group was characterized by a particular combination
of  emission  control   technologies.    The technologies  used  to
separate normal emitters  into  technology groups  were expected to
cause   vehicles   to   respond   differently   to  specific   fuel
modifications or  otherwise  influence  exhaust  emissions.    The
defining characteristics  of each of these nine technology groups is
given in Table III-4.

          Table  III-4:  Normal Emitter Technology Groups
Technology
group
I
2
3
4
5
6
7
8
9
Fuel
system
Multi
Multi
TBI
Multi
Multi
TBI
TBI
TBI
Garb
Catalyst
type
3 -way
3 -way
3 -way
3 -way + Ox
3 -way
3 -way
3 -way + Ox
3 -way
3 -way + Ox
Fuel/Air
system
CLL/No air
CLL/No air
CLL/No air
CLL/Air
CLL/Air
CLL/Air
CLL/Air
CLL/No air
CLL/Air
EGR?
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
For  higher  emitting  vehicles,  EPA  has  found  technology  type
distinctions to  be superfluous  for  reasons outlined  in  Section
IV.A.  Thus all light-duty vehicles fall  into one of ten different
vehicle groups:  either one of the nine normal emitter technology
groups, or the higher emitter group.   The statistical regressions
that derived relationships between  fuel parameters  and  exhaust
emissions made distinctions between these ten vehicle categories,
and so produced separate models  (equations) for each.  Since each
of these  ten equations can  give different emission  results  for
identical fuels,  their results must be combined in such a way as to
                                70

-------
reflect the  contribution of each vehicle  group to the overall  in-
use emissions  of  1990 vehicles.

     The contribution of each of the nine technology groups given
in Table  III-4 to the  total  mobile-source  emissions inventory  is
determined by  two factors:   the fraction of the  1990 sales fleet
accounted for  by each  type of  vehicle  (its  sales weighting) ,  and
the average  emissions for each  vehicle type.  The sales weightings
were derived from reported 1990 light-duty vehicle and truck sales
reports submitted by vehicle manufacturers, and  are equal to  the
fraction  of  1990   vehicles   with   the   corresponding  vehicle
technology40.  Vehicle  sales  fractions  were used  to estimate  the
fraction of  total vehicle miles travelled by  vehicles  of a given
technology type.   To compute  the contribution of each technology
group  to  total emissions,  the  sales  weightings  were adjusted  to
account for  differences in the  average  level of total hydrocarbon
and  NOx emissions   from vehicles  in  each  technology  group,   as
determined through the  EPA Emission Factor program41.   The  sales  and
technology group  emissions weightings are given in Table III-5.

 Table III-5:  Sales  Weightings  and Emissions by Technology Group
Technology
group
1
2
3
4
5
6
7
8
9
Sales
weighting
0.2631
0.2871
0.2181
0.1143
0.0431
0.0053
0.0525
0.0
0.0165
Average
g/mi of
HC
0.278
0.228
0.230
0.435
0.179
0.200
0.389
	
0.258
Average
g/mi of
NOx
0.519
0.570
0.479
0.918
0.389
0.460
0.613
	
0.712
     40
         See EPA Memorandum from Joann Jackson-Stephens to Michael Sklar, "Re-
analysis of 1990  Sales Fleet (CAFE)," March 1, 1993.  Values from this memo
were normalized,  since 1.41% of the  1990 vehicle sales did not  fall into any of
the nine technology groups.
     41
         See EPA Memorandum from Christian E. Lindhjem to Richard A. Rykowski,
"Effect of Oxygenates on Emissions," January 7, 1992.
                                 71

-------
Note  that  technology  group #8  has a  sales  weighting of  0.0%,
indicating that no  vehicles with that particular  technology type
were sold in 1990  in the U.S.  For this reason, technology group #8
will be dropped from subsequent discussions, resulting in a set of
eight normal emitter technology  groups numbered  1,  2,  3,  4,  5,  6,
7,  and 9.

     Within the normal  emitter  class, the relative weighting of the
emission results  from  each technology group is  derived  from the
combination of the  sales weightings and the average emissions for
that technology group.   The relative weighting for each technology
group, which applies only to normal emitting vehicles,  was derived
from the following  equation:
Technology group =
     weighting
	Sales weighting * average emissions for the group
S (Sales weighting *  technology group emission weighting)
This equation gives the values shown  in Table  III-6.

   Table  III-6:  Technology Group Weightings for Normal Emitters
Technology group
1
2
3
4
5
6
7
9
Total
VOC weighting
0.26896
0.24071
0.18446
0.18284
0.02837
0.00390
0.07510
0.01565
0.99999
NOx weighting
0.23842
0.28573
0.18241
0.18321
0.02927
0.00426
0.05619
0.02051
1.00000
     As  stated  before,  a  given  vehicle can  be  categorized as  a
normal emitter or a higher emitter.  Since the weightings shown in
Table III-6 apply only to normal  emitters,  a weighting scheme that
applies  to  all  vehicle  groups'   equations must   combine  the
weightings from Table III-6 with the in-use proportion of emissions
attributable to  each emitter class.   These proportions  depend on
the  inspection  and  maintenance  (I&M)   scenario  assumed for  each
Phase  of the  reformulated  gasoline program.   The  fraction  of
emissions  attributable  to  normal and  higher emitters  for  both
Phases are given in  Table  III-7.
                                72

-------
  Table III-7:  Emitter Class Weightings By Pollutant and Phase

Normal emitters
Higher emitters
Phase I
VOC
0.52
0.48
NOx
0.82
0.18
Phase II
VOC
0.444
0.556
NOx
0.738
0.262
The final weightings that are  applied to the performance estimates
for each  of  the nine  separate  vehicle group equations  for each
exhaust pollutant are  shown in Table  III-8.  The  normal emitter
weightings are calculated from the product of the technology group-
specific weightings  (Table III-6)  and the normal emitter fraction
of total emissions (Table III-7).  Since the  five toxic species of
interest are also classified  as VOCs,  EPA has used the final VOC
weightings to  weight the individual vehicle  group equations for
each exhaust toxic pollutant as well.
   Table III-8: Vehicle Group Weightings for the Complex Model
Vehicle group
Normals, technology
group #1
Normals, technology
group #2
Normals, technology
group #3
Normals, technology
group #4
Normals, technology
group #5
Normals, technology
group #6
Normals, technology
group #7
Normals, technology
group #9
Higher emitters
Phase I
VOC
0.13986
0.12517
0.09592
0.09508
0.01475
0.00203
0.03905
0.00814
0.48000
NOx
0.19550
0.23430
0.14958
0.15023
0.02400
0.00349
0.04608
0.01682
0.18000
Phase II
VOC
0.11942
0.10688
0.08190
0.08118
0.01260
0.00173
0.03334
0.00695
0.55600
NOx
0.17595
0.21087
0.13462
0.13521
0.02160
0.00314
0.04147
0.01514
0.26200
                                73

-------
5.   Summer Baseline Emissions

     The MOBILE model was used as the basis for estimating fleet-
average gram/mile emissions from  summer  baseline  fuel.   However,
the MOBILE model  is equipped only to provide emission estimates for
VOC and NOx.  Therefore, baseline emissions for toxics were treated
separately.  Accordingly,  the  first  section below will present the
treatment of VOC and NOx  baseline emissions,  while the following
section  will present  the treatment  of  baseline emissions  for
toxics.  A table giving the summer baseline emissions for VOC, NOx,
and toxics is presented at the end of Section B.5.b.


a.   VOC and NOx

     To estimate gram/mile emissions  of VOC  and  NOx  from summer
baseline  fuel, only  the  emissions from  1990  model year vehicles
could be considered.  To implement this criterion while maintaining
a focus on in-use emissions from 1990 vehicles  over  their lifetime,
the MOBILE  model was  run in  the year  2015  with all  post-1990
vehicle programs  turned  off   (Tier  1  standards are not  included
because they  do  not  apply to  baseline  1990  vehicles) .   For the
summer season, the daily  low and high  temperatures  modeled were 69
and  94°F, respectively,  for  Class  B   areas  and  72  and  92°F,
respectively, for  Class  C areas  (see Section B.I above).   The
conditions under  which the baseline emission values  were determined
for Phases I and II are shown below:

Phase I:  •    July 11,  1991  version of MOBILE4.1
          •    Light-duty vehicles  (i.e., passenger cars)  only
          •    Basic I/M  program consisting  of  a 2200 rpm idle
               test and a gas cap inspection

Phase II: •    MOBILE5a
          •    Emissions from  light-duty vehicles  and light-duty
               trucks  were   combined   by   weighting   per-mile
               emissions  by  the  percentage  of  vehicle  miles
               traveled
          •    Inspection and maintenance program met EPA's recent
               guidelines for  enhanced  I/M,   including  purge and
               pressure checks

The Agency designed the Phase I conditions to be identical to the
conditions used for the simple model  for the  reasons  outlined in
Section 4 of the preamble.  The MOBILE model  input file for Phase
I baseline emission estimates is given in Appendix  III-A, while the
input file for Phase  II  baseline emission estimates  is given in
Appendix  III-B.

     Although the MOBILE  model  provides gram/mile estimates of VOC
for both  exhaust  and nonexhaust,  the Agency  determined baseline
emissions slightly differently for these two emission categories.

                               74

-------
The reason for this is that the nonexhaust equations in the Complex
Model are  based  on  a simple  regression  using emission estimates
from the MOBILE model for various RVPs.42  The MOBILE  model output
is given  in only two decimal  places.   Thus  when the nonexhaust
regression equations are evaluated for an RVP of  8.7  psi  (the RVP
of summer baseline gasoline),  the resulting emission  estimate for
nonexhaust  VOC  is  slightly  larger  than  the  estimate  provided
directly by the  MOBILE  model  (e.g.  0.492  g/mi vs. 0.483 g/mi for
Phase II,  Class C).   The Agency has determined that the regression
equations are more representative of  the true  relationship between
VOC  emissions  and RVP  than the  data  derived  directly  from the
MOBILE model because of  the error introduced in the  MOBILE model
output through rounding.   Therefore,  the nonexhaust  VOC baseline
values were derived  from regressions based on the emissions from
the MOBILE model instead of from the  MOBILE model  output directly.
The  exhaust  VOC baseline  values,  on the other  hand, were taken
directly from  the MOBILE model output since  MOBILE provided the
only source of baseline exhaust emission estimates.

     As discussed in the  February  1993  NPRM  [58  FR  17175] ,  the
definition of ozone-forming VOCs was  changed  to  match  EPA guidance
that  all   VOC emission estimates  be  in terms  of  nonmethane,
nonethane hydrocarbon emissions.  The official version  of MOBILE4.1
and  MOBILESa  both provide VOC  estimates in  this form directly.
However,  the July 11, 1991 version of MOBILE4.1  (the  MOBILE model
used  to   develop the   simple  model  and  the  Phase   I  baseline
emissions)  gives VOC as nonmethane hydrocarbons only.  When using
VOC  values from the interim  version  of  MOBILE4.1,  then,  it is
necessary to back out the fraction of ethane.   Based on an analysis
of the  Complex  Model  database, the  Agency  determined  that  the
average  ethane  content  was  approximately  2.95%  of  exhaust
nonmethane hydrocarbon emissions.  This value  would be expected to
change as changes are made to the Complex Model database, but not
to such a  degree that  the Phase  I  reformulated gasoline program
would be affected.  Thus  the VOC baseline  emission values  for Phase
I  were  reduced  from  0.46 g/mi nonmethane  VOC  to   0.446  g/mi
nonmethane, nonethane VOC by subtracting  2.95% from the nonmethane
baseline emission estimate.

     The baseline emissions for VOC and NOx  as derived from the
MOBILE model are summarized in Table III-9.
     42
        See Section IV.B below for details on the derivation of the
nonexhaust model for VOC


                                75

-------
     Table  III-9: Summer Baseline Emissions  for VOC and NOx

Exhaust VOC, g/mi
Nonexhaust VOC, g/mi
NOx, g/mi
Phase I
Class B
0.446
0.860
0.660
Class C
0.446
0.769
0.660
Phase II
Class B
0.907
0.559
1.340
Class C
0.907
0.492
1.340
     In addition to the fleet-average baseline emissions,  emitter
class-specific baseline emissions are also needed to calculate the
overall  emissions  performance of  reformulated and  conventional
fuels.   To derive  emitter  class-specific baseline  emissions,  the
fleet and emission  inventories must first be apportioned by emitter
class.   The database on which  the  MOBILE model  is  based provided
this information.   No statistically significant differences in NOx
emissions between normal and higher emitters were found.  The fleet
and emission inventories are  shown in Tables 10 and 11 for Phase I
and Phase II,  respectively.

    Table  111-10: Phase  I Summer  Fleet and Emissions  Inventory

Fraction of in-use fleet
Fraction of VOC inventory
Exhaust VOC emissions, g/mi/car
Fraction of NOx inventory
NOx emissions, g/mi/car
Normal
Emitters
0.82
0.52
0.283
0.82
0.66
High
Emitters
0.18
0.48
1.189
0.18
0.66
Overall
Fleet
1.00
1.00
0.446
1.00
0.66
   Table III-ll: Phase II Summer Fleet and Emissions Inventory

Fraction of in-use fleet
Fraction of VOC inventory
Exhaust VOC emissions, g/mi/car
Fraction of NOx inventory
NOx emissions, g/mi/car
Normal
Emitters
0.836
0.444
0.482
0.82
1.34
High
Emitters
0.164
0.556
3.075
0.18
1.34
Overall
Fleet
1.00
1.00
0.907
1.00
1.34
                                76

-------
     VOC baseline emissions for each emitter class were determined
by  using  the  following  equation,   as  proposed  in  the  draft
Regulatory Impact Analysis:

                         EFI x ABE = EFF x EBE

where,

ABE = Baseline emissions for the fleet (g/mile)
EBE = Baseline emissions for an individual emitter class (g/mile)
EFF = Emitter class fraction of vehicles in the fleet
EFI = Emitter class fraction of the emissions inventory

ABE, EFF, and EFI are derived from the MOBILE model database, and
are given in Tables 10 and 11 above.


b.   Toxics

     As  stated previously,  the MOBILE  model  does  not  provide
emission  estimates  for  toxic compounds.    Thus  the  baseline
emissions for toxics must come from another source.  In addition,
toxic compounds must be  treated separately when estimating exhaust
and nonexhaust  baseline emissions,  since exhaust  baseline toxic
values are derived  mainly  from the Complex  Model database, while
nonexhaust baseline  toxic  values  are computed  directly  from the
nonexhaust Complex Model in conjunction with General Motor's vapor
model.  Baseline emissions of exhaust POM are also treated uniquely
due to the lack of  POM  emissions data.   Baseline levels of these
three types  of  summer toxics emissions  (exhaust other than POM,
nonexhaust, and POM) will be discussed separately in this section.

     The Agency has determined that of  the  five  toxic compounds
that  must  be  regulated in  the reformulated gasoline  and anti-
dumping  programs,  benzene  is the   only  one  that  appears  in
nonexhaust emissions since it is the only regulated toxic compound
naturally occurring in gasoline.  As described in Section IV.B, the
nonexhaust benzene portion of the Complex Model is derived from the
General  Motors  (GM)  vapor  equilibrium model.    This GM model
estimates the  fraction  of benzene  in the vapors above a  pool of
fuel  (conditions  simulating a vehicle's fuel tank).   The Agency
executed a linear regression on the output from the GM model, and
combined the  resulting regression equations with the nonexhaust VOC
equations.   Thus  the  Agency converted benzene fraction estimates
into  a  model  that  provides benzene   emission   estimates  in
grams/mile.   For details  on  the  development of  the  nonexhaust
model, see Section IV.B below.

     Since the nonexhaust VOC model is based on emission estimates
produced from the  MOBILE model, the  nonexhaust  benzene equations in
the Complex Model are the most appropriate source for determining
baseline emissions  of benzene.  Deriving the nonexhaust  benzene

                                77

-------
baseline emissions from the Complex Model is also consistent with
the method described above for determining baseline emissions for
nonexhaust VOC.   Thus  the summer baseline  fuel was  simply run
through the  nonexhaust  benzene portion  of  the  Complex  Model  to
determine  the baseline  emissions of  nonexhaust  benzene.    The
results are shown in Table 111-12:

   Table 111-12:   Summer Nonexhaust Benzene Baseline Emissions

Phase I
Phase II
Class B
9.659
6.241
Class C
8.633
5.505
     To determine baseline emissions for exhaust toxics, the most
complete source of information is  the Complex Model database.  The
Agency therefore performed an analysis of the toxic emissions data
in the Complex Model database to determine the average emissions of
each toxic pollutant.  In this analysis,  only the observations used
in  the development  of  the   Complex  Model equations   for  toxic
compounds were considered.  Normal emitter averages were estimated
separately from higher emitter averages.  Also, the normal emitter
averages were derived  separately for each technology group to mimic
the development of the Complex Model for normal emitters.  For the
same reason,  technology  group distinctions were not made  in the
analysis of  the  higher emitter averages.   Average VOC emissions
were also  calculated  for the same observations.   The  results of
this analysis  for normal emitters are shown in Tables  13 through
16, while the higher emitter results are shown in Table 111-17.
                                78

-------
Table 111-13:   Exhaust  Benzene  and  VOC   Emissions   for Normal
               Emitters
TECHNOLOGY
TYPE
1
2
3
4
5
6
7
9
NUMBER OF
OBSERVATIONS
598
431
300
138
22
70
21
8
EXHAUST
EMISSION
VARIABLE
BENZENE
VOC
BENZENE
VOC
BENZENE
VOC
BENZENE
VOC
BENZENE
VOC
BENZENE
VOC
BENZENE
VOC
BENZENE
VOC
MEAN
EMISSIONS
(mg/mile)
9.724
193.0
11.11
229.7
10.73
199.5
8.086
242 .9
12 .82
322 .4
27.58
440.7
9.492
235.8
6.449
112 .3
RATIO*
0.0504
0.0484
0.0538
0.0333
0.0398
0.0626
0.0402
0.0574
*RATIO  = MEAN EXHAUST BENZENE EMISSIONS/MEAN EXHAUST VOC EMISSIONS
                               79

-------
Table 111-14:  Exhaust 1,3-Butadiene and VOC Emissions for Normal
               Emitters
TECHNOLOGY
TYPE
1
2
3
4
5
6
7
9
NUMBER OF
OBSERVATIONS
574
428
277
132
21
70
21
5
EXHAUST
EMISSION
VARIABLE
1,3-BD
VOC
1,3-BD
VOC
1,3-BD
VOC
1,3-BD
VOC
1,3-BD
VOC
1,3-BD
VOC
1,3-BD
VOC
1,3-BD
VOC
MEAN
EMISSIONS
(mg/mile)
1.225
191.8
1.599
230.2
1.056
206.1
0.868
239.2
1.951
319.4
2 .154
440.7
0.791
235.8
0.824
109.7
RATIO*
0.00639
0.00695
0.00512
0.00363
0.00611
0.00489
0.00336
0.00751
*RATIO  =   MEAN  EXHAUST 1,3-BUTADIENE  EMISSIONS/MEAN EXHAUST VOC
          EMISSIONS
                                80

-------
Table Ill-IB:  Exhaust Acetaldehyde  and VOC Emissions for Normal
               Emitters
TECHNOLOGY
TYPE
1
2
3
4
5
6
7
9
NUMBER OF
OBSERVATIONS
528
427
276
134
21
66
21
8
EXHAUST
EMISSION
VARIABLE
ACET
VOC
ACET
VOC
ACET
VOC
ACET
VOC
ACET
VOC
ACET
VOC
ACET
VOC
ACET
VOC
MEAN
EMISSIONS
(mg/mile)
0.9725
187.4
1.225
230.5
1.263
199.4
1.638
241.0
1.014
322 .2
2.883
440.0
1.019
235.8
0.9363
112 .3
RATIO*
0.00519
0.00531
0.00633
0.00680
0.00315
0.00655
0.00432
0.00834
*RATIO  =   MEAN  EXHAUST  ACETALDEHYDE  EMISSIONS/MEAN EXHAUST VOC
          EMISSIONS
                                81

-------
Table 111-16:  Exhaust Formaldehyde  and VOC Emissions for Normal
               Emitters
TECHNOLOGY
TYPE
1
2
3
4
5
6
7
9
NUMBER OF
OBSERVATIONS
529
425
277
135
21
70
21
8
EXHAUST
EMISSION
VARIABLE
FORM
VOC
FORM
VOC
FORM
VOC
FORM
VOC
FORM
VOC
FORM
VOC
FORM
VOC
FORM
VOC
MEAN
EMISSIONS
(mg/mile)
2.336
187.4
2 .727
230.3
3.037
199.8
5.715
240.7
2.438
322 .2
6.431
440.7
2 .754
235.8
3.091
112 .3
RATIO*
0.0125
0.0118
0.0152
0.0237
0.00757
0.0146
0.0117
0.0275
*RATIO  =   MEAN  EXHAUST  FORMALDEHYDE  EMISSIONS/MEAN EXHAUST VOC
          EMISSIONS
                                82

-------
Table 111-17:  Average Toxic and VOC Emissions from Higher Emitters




Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde


Number of
observations
269
270
269
269
Average
toxic
emissions,
mg/mi
95.05
23.85
11.45
27.60
Average
VOC
emissions,
mg/mi
2082
2117
2117
2116



Ratio*
0.0457
0.0113
0.00541
0.0130
*RATIO =  MEAN EXHAUST TOXIC EMISSIONS/MEAN EXHAUST VOC EMISSIONS


     The  toxics  emission  inventory  is  expected  to be  roughly
proportional to the VOC inventory because toxics are by definition
volatile   organic   compounds   (though  the   effects  of   fuel
modifications on toxics emissions are expected to differ from the
effects of the same fuel modifications on exhaust VOC emissions).
Therefore, ratios of the average emissions of each toxic pollutant
with  respect  to  the  average emissions  of  VOCs  for  the  same
observations  were  also  calculated.   These  ratios were used  to
compensate for the fact that  the observations which contributed to
the average values  in  Tables  13-17  were  not  based only on summer
baseline fuel and were  not based on vehicles with representative
in-use VOC emission levels.   Summer baseline emission values for
exhaust toxics were calculated using the following procedure:

     1.   The normal emitter toxic  emission ratios from Tables 13
          through 16 were weighted by the  normal emitter technology
          group weighting factors.   The normal emitter technology
          group  weighting  factors   were  derived  in  Subsection
          III.E.4  above,  and  are  given  in  Table  III-6.    They
          represent the  fraction of the mobile  source  emissions
          inventory  for  normal  emitters  that  comes from  each
          technology group.

     2.   The weighted ratios from step (1)  were summed to obtain
          a single,  fleet-average ratio for each toxic  emission for
          normal emitters  as shown  in Table 111-18.   This  step
          yields overall  toxic-to-VOC  ratios  for normal emitters
          that are  analogous  to  the values  presented for  higher
          emitters in Table 111-17.
                                83

-------
   Table  111-18:
Average Toxic and VOC Emissions from Normal
          Emitters

Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
Number of
observations
1588
1528
1481
1486
Ratio*
0.0465
0.00556
0.00566
0.0149
*RATIO =  MEAN EXHAUST TOXIC EMISSIONS/MEAN EXHAUST VOC EMISSIONS
     3.    The ratios  in Tables  17  and 18 are  based on  all  the
          available data in the  Complex Model database  for which
          both toxics and NMHC data were  available.   Since these
          ratios  are  not  based  strictly  on   summer  baseline
          gasoline,   they  cannot  provide   the   toxic   baseline
          emissions  for  summer  baseline  fuel.    To  convert  the
          ratios from Tables 17 and 18  into ratios that reflect the
          effect of summer baseline fuel  on  toxic  emissions when
          used in representative in-use 1990 vehicles, a correction
          was made.   This  correction was carried out in two steps.
          The first  step  corrected for the  fact that  the toxic
          emission  averages  are  not  based  on  summer  baseline
          gasoline.   The second  step  corrected  for  the  fact that
          the  VOC   emission  averages  are not  based  on  summer
          baseline gasoline.  These two corrections were performed
          sequentially:

     3.1)  The average value of  each fuel  parameter was determined
          for the same observations that were  used to determine the
          average toxics  and VOC values  in Table 111-17  and  18.
          This step was performed separately for the normal  and
          higher emitter observations, and for  each  toxic.   Thus
          there were a total of eight  "average" fuel descriptions.
          Discrepancies  between  these average  fuels and  summer
          baseline fuel  caused discrepancies between the ratios in
          Tables 17  and  18  and the true ratios for summer baseline
          gasoline.   The average  fuels for each toxic are given in
          Tables 19 and  20.
                               84

-------
   Table  111-19:  Average Fuels  for Normal Emitters

MTBE, wt% oxygen
ETBE, wt% oxygen
Ethanol, wt% oxygen
TAME, wt% oxygen
Sulfur, ppm
RVP, psi
E200, percent evap
E300, percent evap
Aromatics, vol%
Olefins, vol%
Benzene, vol%
Benzene
1.236
0.021
0.198
0.015
218.6
8.264
47.55
84.94
28.58
7.943
0.953
1,3-
Butadiene
1.254
0.022
0.199
0.014
220.4
8.268
47.49
84.90
28.41
7.969
0.921
Acet-
aldehyde
1.311
0.023
0.206
0.013
218.3
8.248
48.43
85.28
28.15
8.028
0.966
Form-
aldehyde
1.313
0.021
0.209
0.013
217.9
8.249
48.43
85.31
28.14
8.005
0.964
   Table  111-20:  Average Fuels  for Higher Emitters
Fuel parameter
MTBE, wt% oxygen
ETBE, wt% oxygen
Ethanol, wt% oxygen
TAME, wt% oxygen
Sulfur, ppm
RVP, psi
E200, percent evap
E300, percent evap
Aromatics, vol%
Olefins, vol%
Benzene, vol%
Benzene
1.457
0.020
0.361
0.009
255.5
7.810
52.49
84.54
26.03
5.699
0.871
1,3-
Butadiene
1.466
0.020
0.346
0.009
255.7
7.803
52.51
84.57
25.98
5.692
0.867
Acet-
aldehyde
1.464
0.010
0.360
0.009
255.7
7.810
52.55
84.60
25.98
5.683
0.868
Form-
aldehyde
1.457
0.020
0.360
0.009
254.6
7.812
52.56
84.62
26.01
5.667
0.867
3.2)  The Complex Model  was used to determine a multiplicative
     factor to convert emissions of each toxic  for the average
     fuel into  emissions  of  each toxic  for  summer baseline
     fuel.    This  step  was accomplished by  evaluating  the
     normal emitter toxic  equations with  the fuels from Table
     111-19,  evaluating the  higher emitter  toxic  equations
     with the fuels  from  Table 111-20,  and  evaluating both
     normal and  higher  emitter toxic  equations  with summer
     baseline fuel.  The  multiplicative factor  can  then be
     written as  follows:

          Toxic  correction factor = exp(f(b))/exp(f(a))
                           85

-------
     where  f  is  the normal  or  higher emitter  polynomial
     equation for the toxic of interest, b is summer baseline
     gasoline,  and a  is  the  average fuel from  either Table
     111-19 or Table  111-20.  The toxic emission correction
     factors are given in Table  111-21.

   Table  111-21:  Toxic  Emission  Correction Factors
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
Normal emitter
correction factor
1.371
1.121
1.011
0.9373
Higher emitter
correction factor
1.500
1.402
0.9970
0.8151
3.3)  The ratios from Tables 17 and 18 were multiplied by the
     appropriate toxic correction factors from Table 111-21.
     This step converted the toxics portion of each ratio to
     a value  representing the  toxic emissions  from summer
     baseline gasoline.

3.4)  Beginning with the same average  fuels, the Complex Model
     was used to determine a multiplicative factor that will
     convert  VOC  emissions  for  the  average fuel  to  VOC
     emissions  for summer  baseline  fuel.    This step  was
     accomplished  by  evaluating  the  normal  emitter  VOC
     equation with the  fuels from Table 111-19, evaluating the
     higher emitter VOC  equation with the fuels  from Table
     III-20, and evaluating both normal and higher emitter VOC
     equations with summer baseline fuel. The multiplicative
     factor can then be written as follows:

          VOC correction factor = exp(g(a))/exp(g(b))

     where g  is  the normal emitter  polynomial equation for
     VOC, a is the average fuel from Tables 19 and 20,  and b
     is  summer  baseline  gasoline.    Note  that  the  VOC
     correction factor in this step has the reciprocal form of
     the  factor  given in  step  (3.2)  because this  step  is
     correcting the denominator of each ratio instead of the
     numerator.  The VOC correction factors are given in Table
     111-22 .
                           86

-------
    Table  111-22:  VOC Emission  Correction  Factors
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
Normal emitter
correction factor
0.8685
0.8694
0.8606
0.8603
Higher emitter
correction factor
0.9045
0.9039
0.9042
0.9044
3.6)  The  ratio  from  step  (3.3)   was  multiplied  by  the
     appropriate  factors  from  Table   111-22.    This  step
     converts  the  VOC  portion  of  each  ratio  to  a  value
     representing  the  VOC  emissions  from summer  baseline
     gasoline.   The  ratios from Tables  17 and  18  were now
     fully corrected to  represent  summer  baseline gasoline,
     and are shown in Table 111-23.

       Table 111-23:   Corrected Toxic:VOC Ratios
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
Normal emitter
ratio
0.0554
0.00542
0.00492
0.0120
Higher
emitter
ratio
0.0620
0.0143
0.00488
0.00962
4.   Each of  the  four normal emitter  toxic  emission ratios
     from Table  111-23 was  multiplied by  the Phase  I  VOC
     baseline for normal emitters,  which is 283 mg/mi  (Table
     111-10).   For the higher emitter ratios, each value from
     Table 111-23 was multiplied by the Phase I VOC baseline
     for higher  emitters,  which is  1189  mg/mi.   This step
     produced the  Phase  I  normal and  higher  emitter toxics
     baselines as shown in Table 111-24.
                           87

-------
Table 111-24:
Phase I Summer Toxic Baseline Emissions by Emitter
               Class
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldeyhde
Formaldehyde
Normal emitter
baseline, mg/mi
15.66
1.533
1.391
3.404
Higher emitter
baseline, mg/mi
73.68
16.98
5.799
11.44
     5.   The  normal  and  higher  emitter  toxic  baselines  were
          weighted by  the fraction of the in-use fleet attributable
          to each  emitter  class,  and the results were  summed to
          obtain the final Phase I toxic baseline emissions.  The
          fraction of the  VOC  inventory for Phase  I  is  given in
          Table 111-10.   The final Phase I exhaust toxic baseline
          emissions for summer are shown in Table 111-25.

     Table  111-25:  Phase  I  Summer  Toxic  Baseline Emissions
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldeyhde
Formaldehyde
Baseline, mg/mi
26.10
4.313
2.185
4.850
     6.   To obtain Phase II  summer baseline emissions for toxics,
          steps 4 and 5  were  repeated using Phase II VOC baselines
          and  fleet  fractions (Table  III-ll) .   The  ratios  from
          Table 111-23 were multiplied by 482  and  3075  mg/mi for
          normal  and  higher  emitters,   respectively,   and  the
          resulting emitter class-specific baseline  emission values
          were combined via  the Phase II in-use fleet fractions of
          0.836  and  0.164   for  normal  and  higher  emitters,
          respectively.   The final Phase II exhaust toxic baseline
          emissions for summer are shown in Table 111-26.
                                88

-------
     Table 111-26:  Phase II Summer Toxic Baseline Emissions
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldeyhde
Formaldehyde
Baseline, mg/mi
53.54
9.382
4.440
9.697
     Data  on polycyclic  organic material  (POM)  were  extremely
limited  since  very few  test  programs  had   the  capability  of
measuring POM.  The  lack of data required EPA to pursue a different
approach to establish baseline emissions for POM.   An analysis of
the limited  data  available  to EPA showed that (1)  POM emissions
were  proportional   to   exhaust  VOC  emissions,   and   (2)   the
proportionality constant  is  0.003355.  Thus the POM baseline can be
determined from the  exhaust  VOC baseline for both Phase  I and II in
the following manner:
Phase I:
Phase II:
POM baseline
POM baseline
POM baseline

POM baseline
POM baseline
POM baseline
0.003355 x exhaust VOC baseline
0.003355 x 0.446 g/mi x 1000 mg/g
1.496 mg/mi

0.003355 x exhaust VOC baseline
0.003355 x 0.907 g/mi x 1000 mg/g
3.043 mg/mi
     The final summer baseline emissions are shown in Table 111-27

       Table 111-27:  Summer Baseline Emissions in mg/mile

Exhaust VOC
Nonexhaust VOC
NOx
Exhaust benzene
Nonexhaust benzene
Acetaldehyde
Formaldehyde
1, 3 -Butadiene
POM
Phase I
Class B
446.00
860.48
660.00
26.10
9.66
2 .19
4.85
4.31
1.50
Class C
446.00
769.10
660.00
26.10
8.64
2 .19
4.85
4.31
1.50
Phase II
Class B
907.00
559.31
1340.00
53.54
6.25
4.44
9.70
9.38
3.04
Class C
907.00
492 .07
1340.00
53.54
5.51
4.44
9.70
9.38
3.04
                                89

-------
6.   Winter Baseline Emissions

     The calculation of winter baseline emissions was approached in
a very  different manner  than used  to estimate  summer  baseline
emissions.    EPA  chose  to  use   a  different  approach  since
insufficient  testing  has been  done  under  winter conditions  to
develop a separate Complex Model for the winter season.  The Agency
determined  that  the  MOBILE  emissions model  provided the  most
appropriate sources of information on the fleet-average emissions
produced when using winter baseline fuel.

     One simplifying factor in the determination of winter baseline
emissions is that,  as  discussed in Section IV.E below,  the Agency
has determined that emissions of nonexhaust VOC and benzene under
winter conditions should be  set  to  zero.  Thus baseline nonexhaust
emissions under winter conditions do not need to be calculated.

     Many aspects of  the derivation  of winter baseline emissions
parallel the derivation of  summer  baseline  emissions.   Thus this
section  will  treat  VOC  and NOx separately  from toxics  for  the
reasons discussed in Section B.5 above.
a.   VOC and NOx

     The  MOBILE  model  was run  in  a  manner  identical to  that
discussed above for the derivation  of summer baseline emissions of
exhaust VOC and NOx, but with two exceptions.   First,  minimum and
maximum wintertime  temperatures  of 39 °F and 57  °F were used in
place  of  the  summer  temperature  ranges.    Second,  the  winter
baseline fuel was used in place of the summer baseline fuel.   The
remaining details of running the MOBILE model to determine  winter
baseline emissions of exhaust VOC and NOx are shown below:

Phase I:  •    July 11, 1991 version of MOBILE4.1
          •    Light-duty vehicles (i.e., passenger cars)  only
          •    Basic  I/M  program consisting  of  a 2200 rpm  idle
               test and a gas cap inspection

Phase II:  •    MOBILE5a
          •    Emissions from  light-duty vehicles and light-duty
               trucks   were  combined   by   weighting   per-mile
               emissions  by  the  percentage  of  vehicle   miles
               traveled
          •    Inspection and maintenance program  met EPA's recent
               guidelines  for  enhanced I/M,   including  purge  and
               pressure checks

These conditions are identical to those used to develop the summer
baseline emission  estimates.   Winter VOC  baseline  emissions  for
Phase   I   were  corrected  to  reflect   nonmethane,   nonethane
hydrocarbons in a manner identical to that used to correct  summer

                                90

-------
VOC baseline emissions.  The  final  winter baseline emissions for
exhaust VOC and NOx are shown in Table III-28.
     Table 111-28: Winter Baseline Emissions for VOC and NOx

Exhaust VOC, g/mi
Nonexhaust VOC, g/mi
NOx, g/mi
Phase I
Class B
0.660
0.000
0.750
Class C
0.660
0.000
0.750
Phase II
Class B
1.341
0.000
1.540
Class C
1.341
0.000
1.540
     In addition to the fleet-average baseline emissions,  emitter
class-specific baseline emissions were  calculated  using the same
method  discussed  in  Section  B.5  above  for  summer  baseline
emissions.  The emitter class  fractions  of  the VOC inventory and
in-use  fleet  are  the  same  in  the  summer  and  the winter,  so
differences  between  summer  and  winter  emitter  class-specific
baseline emissions are derived solely from  differences  in fleet-
average baseline emissions.  As with the  summer baseline emissions,
no statistically significant differences in NOx emissions between
normal and  higher  emitters were  found.   The  fleet  and emission
inventories for winter are shown  in  Tables  29  and  30 for Phase I
and Phase II,  respectively.

   Table  111-29: Phase  I Winter Fleet and Emissions  Inventory

Fraction of in-use fleet
Fraction of VOC inventory
Exhaust VOC emissions, g/mi/car
Fraction of NOx inventory
NOx emissions, g/mi/car
Normal
Emitters
0.82
0.52
0.419
0.82
0.75
Higher
Emitters
0.18
0.48
1.760
0.18
0.75
Overall
Fleet
1.00
1.00
0.660
1.00
0.75
                                91

-------
   Table 111-30: Phase II Winter Fleet and Emissions Inventory

Fraction of in-use fleet
Fraction of VOC inventory
Exhaust VOC emissions, g/mi/car
Fraction of NOx inventory
NOx emissions, g/mi/car
Normal
Emitters
0.836
0.444
0.712
0.82
1.54
Higher
Emitters
0.164
0.556
4.546
0.18
1.54
Overall
Fleet
1.00
1.00
1.341
1.00
1.54
     VOC baseline emissions for each emitter class were determined
using the same relationship described in Section B.5 above:
where,

ABE =
EBE =
EFF =
EFI =
                         EFI x ABE = EFF x EBE
 Baseline emissions for the fleet (g/mile)
 Baseline emissions for an individual emitter class (g/mile)
 Emitter class fraction of vehicles in the fleet
 Emitter class fraction of the emissions inventory
ABE, EFF, and EFI are derived from the MOBILE model database, and
are given in Tables 29 and 30 above.
b.
Toxics
     As  stated  previously,   the   Agency  has  determined  that
nonexhaust emissions are negligible in the winter.  Therefore only
exhaust  toxic  baseline  emissions  need  be  determined.    Winter
emissions of benzene, 1,3-butadiene, acetaldehyde, and formaldehyde
are discussed  first,  followed by  a  brief discussion of POM.   A
table giving all  the winter baseline  emissions  is presented at the
end of this subsection.

     The steps in the derivation of winter baseline emissions for
exhaust toxics parallel  those presented in Section B.S.b above for
summer baseline  toxic emissions.   Indeed, steps  (1)  and  (2)  are
identical  in  both  cases.   Deviations from the  summer  baseline
emission derivation procedure begin  in  step 3 and  are  detailed
below:

     3.   Since  the  ratios  in Tables 17  and 18 represent  average
          fuels in the Complex Model  database,  they cannot provide
          the toxic  baseline  emissions for  winter baseline fuel.
          The  conversion of  the   ratios  from Tables  17   and  18
                                92

-------
          required the same six steps  that  are  described for the
          summer scenario (i.e.  steps  3.1 to 3.6), with the single
          exception  that winter  baseline  fuel replaced  summer
          baseline fuel in all calculations.  The final, corrected
          ratios representing winter baseline gasoline are given in
          Table III-31.

      Table  111-31:   Corrected  Toxic:VOC Ratios for  Winter
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
Normal emitter
ratio
0.0509
0.00601
0.00544
0.0134
Higher
emitter
ratio
0.0635
0.0164
0.00538
0.00983
     4.   Each of  the  four normal emitter  toxic  emission ratios
          from Table 111-31 were multiplied by the Phase I winter
          VOC baseline  for normal emitters,  which is  419  mg/mi
          (Table 111-29) .   For  the  higher emitter ratios,  each
          value from Table  111-31 was multiplied by  the  Phase I
          winter VOC baseline for higher  emitters,  which  is 1760
          mg/mi.    This  step produced  Phase I normal  and higher
          emitter baselines as shown in Table 111-32.

Table 111-32:  Phase I Winter Toxic Baseline Emissions by Emitter
                              Class
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldeyhde
Formaldehyde
Normal emitter
baseline, mg/mi
21.29
2.515
2 .277
5.625
Higher emitter
baseline, mg/mi
111.7
28.95
9.463
17.31
     5.   The  normal  and  higher  emitter  toxic  baselines  were
          weighted by  the fraction of the in-use fleet attributable
          to each  emitter  class,  and the results were  summed to
          obtain the final Phase I toxic baseline emissions.   The
          fraction of the  VOC  inventory for Phase I  is  given in
          Table 111-29.   The final Phase I exhaust toxic baseline
          emissions for summer are shown in Table 111-33.
                                93

-------
      Table  111-33:   Phase  I Winter  Toxic  Baseline  Emissions
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldeyhde
Formaldehyde
Baseline, mg/mi
37.57
7.273
3.571
7.728
     6.   To obtain Phase II  winter baseline emissions for toxics,
          steps 4 and 5  were  repeated using Phase II VOC baselines
          and fleet fractions (Table 111-30).   Thus  the ratios from
          Table 111-31 were  multiplied by 712  and  4546  mg/mi for
          normal  and  higher  emitters,   respectively,   and  the
          resulting emitter class-specific baseline  emission values
          were weighted via the Phase II in-use fleet fractions of
          0.836 for normal emitters and 0.164 for higher emitters.
          The final Phase II exhaust toxic baseline emissions for
          winter are shown in Table 111-34.

     Table 111-34:  Phase II Winter Toxic Baseline Emissions
Toxic pollutant
Benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
Baseline, mg/mi
77.62
15.84
7.248
15.34
     The winter POM baseline can be determined from the exhaust VOC
baseline for both Phase I and II with the following equations:
Phase I:
Phase II:
POM baseline
POM baseline
POM baseline

POM baseline
POM baseline
POM baseline
0.003355 x exhaust VOC baseline
0.003355 x 0.660 g/mi x 1000 mg/g
2.214 mg/mi

0.003355 x exhaust VOC baseline
0.003355 x 1.341 g/mi x 1000 mg/g
4.499 mg/mi
     The final winter baseline emissions are shown in Table 111-35.
                                94

-------
       Table 111-35:  Winter Baseline Emissions in mg/mile

Exhaust VOC
Nonexhaust VOC
NOx
Exhaust benzene
Nonexhaust benzene
Acetaldehyde
Formaldehyde
1, 3 -Butadiene
POM
Phase I
660.00
0.00
750.00
37.57
0.00
3.571
7.728
7.273
2 .214
Phase II
1341.00
0.00
1540.00
77.62
0.00
7.248
15.34
15.84
4.499
7.   Response to Comments

     A number of comments have been received concerning the use of
the MOBILE model for determining both  the  baseline emissions for
VOC  and NOx  and  the  nonexhaust  VOC  model.    Some  commenters
questioned the appropriateness of using MOBILE4.1 in place of the
more  recent  MOBILESa.    The  Agency  continues  to believe  that
retention of a  MOBILE4.1-derived baseline  maintains  a consistent
baseline and exhaust:nonexhaust  ratio  throughout  Phase  I.   This
consistency also allows  refiners the  lead time they  require  in
order to comply with baseline changes when they reach Phase II in
2000.  These points are discussed in more detail in  Section B above
and in Section III  of the preamble.

     Other commenters asserted that the MOBILE models have become
an  integral  part  of  the  RFG   rulemaking process,  and  should
therefore be subject to  a notice and comment  period.   The MOBILE
models are publicly available, and the  EPA has considered comments
on  their  results,  as  used in  this  rulemaking,  to  be  valid.
However, EPA does not consider the MOBILE models themselves to be
subject  to notice  and comment provisions  since they do  not,  by
themselves, constitute a rule.  EPA has conducted public workshops
on the various iterations of the MOBILE model,  including MOBILES,
at which it solicited suggestions and input from the public.   The
resulting models represent  EPA's understanding of the  impact  of
various  conditions  and policy options  on emissions  from mobile
sources; since  the Agency does not have  access  to a  superior source
of such  information, the Agency  has chosen to  utilize the MOBILE
models to  help  analyze the  impact of  its  rules on mobile source
emissions.
                                95

-------
C.   Appendices
Appendix III-A:     MOBILE4.1  Input  File  Used  to  Develop the
                    Simple Model  and the  Phase I  Complex  Model
                    (must  be   run  with   interim   version   of
                    MOBILE4.1)

Appendix III-B:     MOBILE5a Input File Used To Develop the Phase
                    II Complex Model
                                96

-------
Appendix III-A:     MOBILE4.1  Input  File  Used  to  Develop the
                    Simple Model  and the  Phase I  Complex  Model
                    (must  be   run  with   interim   version   of
                    MOBILE4.1)

1          PROMPT -
MOBILE4.1 v.7/11/91: Simple Model correlations
1          TAMFLG -
1          SPDFLG -
1          VMFLAG -
1          MYMRFG -
1          NEWFLG -
2          IMFLAG - enter I/M program
1          ALHFLG -
2          ATPFLG - enter ATP
2          RLFLAG - Stage II refueling controls
2          LOCFLG - enter LAP record once
1          TEMFLG -
3          OUTFMT - print 112 column descriptive output format
4          PRTFLG - print exhaust HC, CO and NOx results
1          IDLFLG -
3          NMHFLG - print VOC
3          HCFLAG - print HC components
83 20 68 20 00 00 100 1 1 2221 2 11 20 20 20
83 68 20 2221 11 100.  11111212
89 1 086 086
RegNeg           C 71.6 91.6 11.5 08.7 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 08.5 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 08.3 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 08.1 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.9 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.7 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.5 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.2 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.0 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 08.7 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 08.5 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.8 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.6 89 1

                                97

-------
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.4 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.2 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.0 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 06.8 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 06.6 89 1
1 15 19.6 75.0 20.6 27.3 20.6
                                98

-------
Appendix III-B:     MOBILE5a Input File Used To Develop the Phase
                    II Complex Model
I          PROMPT -
Phase II Complex model baseline runs for LDGVs & LDGTs
I          TAMFLG -
1          SPDFLG -
1          VMFLAG -
1          MYMRFG -
5          NEWFLG - no CAA mandates are modeled
2          IMFLAG - enter I/M program
1          ALHFLG -
8          ATPFLG - enter ATP; include pressure and purge checks
2          RLFLAG - enter Stage II
1          LOCFLG -
1          TEMFLG -
3          OUTFMT - print 112 column descriptive output format
1          PRTFLG - print exhaust & evap HC only
1          IDLFLG -
3          NMHFLG - print VOC
2          HCFLAG - no HC components
83 20 68 20 03 03 096 1 1 2221 4211 0.80 20.0 2.00 Enhanced IM
83 84 20 2221 11 096. 12211111                     ATP
83 83 20 2221 11 096.                              Pressure test
83 86 20 2221 11 096.                              Purge test
89 1 80. 60.                                       Stage II
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 06.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 06.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 06.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1

                                99

-------
	  B 69.4 94.0 09.5 08.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 06.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 06.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 06.5 90 1 1 1
                               100

-------
IV.  Complex Model

     The Complex Model is designed to predict emissions as a
function of fuel properties.  An empirical model is more
accurate, faster, cheaper and more practical to use than actual
vehicle testing in predicting the emission effects of fuel
modifications.  Implicit in Section 211(k) the Clean Air Act,  and
confirmed by actual test data, is the assumption that emissions
from vehicles can be linked to the properties of gasolines.  The
vehicle itself has been found to be the single most significant
determinant of the level of emissions from any given vehicle;  to
identify the effects of fuel property modifications on emissions,
EPA found it necessary to separate the emissions effect of each
vehicle from fuels' effects.  The purpose of the Complex Models
developed in the following section is to be able to predict
emissions based on fuel properties which are readily measurable
and easily controllable.

     EPA has developed separate complex models for exhaust and
nonexhaust emissions which are based on entirely different data
sources.  While the exhaust model is based on data from a number
of different test programs, the nonexhaust model is based on data
generated from EPA's ongoing nonexhaust emissions testing program
that has been used to develop EPA's MOBILE emission inventory
models.  EPA believes this data to be sufficient to model the
relationship between fuel properties and nonexhaust VOC emissions
since nonexhaust emissions are better understood from a
theoretical standpoint and are influenced by fewer fuel
properties than are exhaust emissions.

     The exhaust Complex Model described in the following
sections has undergone some changes since it was proposed in the
February 1993 NPRM.  These changes have been made in response to
the following factors:

     •    EPA's use of more appropriate data analysis methods
     •    EPA's improved understanding of the relationship
          between fuel characteristics and emissions
     •    Comments received in response to the February NPRM,
          public workshops, and EPA's docket submission that
          described a number of alternative exhaust complex
          models.

The Complex Model described below was developed through an open,
iterative process.  EPA conducted six separate public workshops
to review various iterations of its models and formally submitted
five previous versions to the docket.  As a result, the Complex
Model has been thoroughly peer reviewed,  and EPA has considered
the suggestions and comments made by various parties seriously in
developing the model.  While complete consensus between all
parties regarding the model and the process used to develop it
                               101

-------
has not been achieved, the remaining differences are decidedly
minor.
A.   Exhaust Emissions Model

     When EPA began the model building process, several key
decisions had to be made regarding the selection of the
appropriate dependent and independent variables, the selection of
appropriate data, the selection of an appropriate modeling
method, and the choice of a simplification methodology.  Each of
these decisions is reviewed in detail in Sections B.I through B.5
below.

     Fuels can be characterized in terms of a number of different
sets of fuel parameters.  EPA used the results of individual fuel
studies, input from its public workshops, preliminary test
results, engineering intuition, available data, and ease of
measurement to select the set of fuel parameters used to model
exhaust emissions.  The Complex Models include the effects of
fuel benzene (for benzene emissions only),  aromatics, olefins,
RVP, E200, E300, sulfur, oxygen content and oxygenate type  (for
acetaldehyde and formaldehyde emissions only) on emissions.
Other parameters were considered by EPA but rejected after
considering the comments received at public workshops and the
available test data.

     In the February 1993 proposal, EPA represented distillation
characteristics by parameters measuring the temperature at which
specified percentages of the fuel would have evaporated,
specifically T50 and T90 (representing the temperatures at which
50% and 90% of the fuel would have evaporated, respectively).   At
the time of that proposal,  EPA indicated that it might replace
these temperature-based parameters with parameters measuring the
percentage of fuel that had evaporated at a specified
temperature, specifically E200 and E300  (representing the
percentage of fuel that would evaporate at 200 degrees Fahrenheit
and 300 degrees Fahrenheit, respectively).   Unlike the
temperature parameters, the percent evaporated parameters for a
mixture of two fuels can be estimated from a weighted average of
the percent evaporated values of its constituent blends, thereby
simplifying calculations for refiners and producers of gasoline.
EPA has chosen to convert its models to a percent evaporated
basis since the NPRM was issued, removing the T50 and T90 terms
from its models in the process. The following correlations can be
used to convert any temperature basis values to percent
evaporated values:

               E200 (%) = 147.91 - (0.49 x T50(in °F) )
               E300 (%) = 155.49 - (0.22 x T90(in °F) )
                               102

-------
These correlations were derived from the Complex Model fuels
database.  A linear regression model was used to relate all
available E200 and E300 values to all corresponding T50 and T90
values in the database.  Other non-linear relationships and
regressions were investigated, but they gave no more accurate
predictions and no greater explanatory power (in terms of the R2
of the correlations)  than the simple linear relationships shown
above.  Thus, EPA chose to use the linear relationships to
convert temperature-based distillation parameters to a percent
evaporated basis.

     Separate models were developed for exhaust VOC, NOx, and
exhaust emissions of the five air toxics specified in Section
211(k), specifically benzene, 1,3-butadiene, formaldehyde,
acetaldehyde and polycyclic organic matter.   Non-methane
hydrocarbon  (NMHC) was used as the VOC model's dependent variable
for all runs; NMHC was converted to VOC by subtracting out the
approximately constant ethane fraction of NMHC (see Sections III
and IV.B.2).

1.    Data Sources

     Since no single study of the effect of fuel modifications on
exhaust emissions was designed to look at effects of all fuel
parameters independently across their full range, EPA chose to
base its exhaust Complex Model on a number of studies.  EPA used
all data submitted as long as the following conditions were met:

     •    The data were based on valid tests.
     •    Data were submitted to EPA in time for development of
          the exhaust models.
     •    Data were based on tests of vehicles classified as
          having  "1990 technology" by the vehicles'
          manufacturer.
     •    No evidence of inaccuracy, error, or
          unrepresentativeness existed in the data.

Each of these points are discussed in more detail in Sections
A.I.a through A.I.e.   Thus, the Complex Models described in this
section are based on data generated from a number of different
exhaust emission testing studies, rather than relying on the
results of a single test program.  These models will be used to
estimate the exhaust emission performance of reformulated and
conventional gasolines based on their fuel properties.

a.    Included Studies

     The Complex Model database for exhaust emissions includes
data from the following studies:
                               103

-------
EPA's Emission Factors (EF) testing program, which is
designed to examine the effect of oxygenates on
emissions.  Valid emissions data acquired through
approximately August 20,  1992 were used from the EF
testing program.

The Auto/Oil Test Program:  This program included a
number of distinct studies sponsored jointly by
companies in the automobile and oil industries.  The
AMOT study  (Auto/Oil-I) was parametrically designed to
examine the effects of aromatics, olefins, T90 and
oxygen (in the form of MTBE)  on emissions.  All
possible combinations of high and low values of each
parameter were tested.  The Phase I and Phase II sulfur
studies (Auto/Oil-II and VIII) were designed to
determine the effects of different sulfur levels on
emissions.  The RVP/Oxygenate study (Auto/Oil-VI) was
designed to isolate the effects of MTBE, ETBE, and
ethanol at different RVP levels on emissions.  The TAME
study  (Auto/Oil-IX) was designed to investigate the
effects on emissions when using TAME as the oxygenate.
Matrix A,  Matrix A2 and Matrix B heavy hydrocarbon
studies which were designed to investigate the effects
of E200 and E300 on exhaust emissions.

The RVP/Oxygenate study sponsored by General Motors,
the California Air Resources Board, and the Western
States Petroleum Association  (GM/CARB/WSPA), which
examined the effects of RVP,  the driveability index
(defined as [T10 + (3 x T50)  + (1.5 x T90)]), and
different oxygenates on emissions.

The RVP/Oxygenate and Aromatics studies sponsored by
the American Petroleum Institute  (API-RVP/OXY and API-
Aromatics).   The API aromatics study was designed to
analyze the effects of different types of aromatics on
emissions; the API-RVP/OXY study was designed to
determine the effect of RVP,  different oxygenates, and
temperature variations on emissions.

Phase I and Phase II of EPA's reformulated gasoline
testing program, performed at Automotive Testing
Laboratories (ATL-I and ATL-II).   ATL-I was designed to
examine the effects of sulfur, RVP, oxygen content, and
T90 on emissions.  ATL-II was designed to study the
effects of oxygen content, oxygenate type, T50, T90,
distillation end point, olefins and heavy end aromatics
on emissions.

ARCO's EC-X study, which was designed to examine the
effects of severe, multi-parameter reformulations
                     104

-------
          (involving changes in fuels' oxygen, sulfur,
          distillation parameter,  aromatics, and olefin levels)
          on emissions.

          Chevron's distillation studies (Chevron studies A-F),
          which examined the effects of T90, T50, T10,
          driveability index, aromatics, and RVP on emissions.

          UNOCAL's reformulated gasoline study (#514C),  which was
          designed to study the effects of T10, T50, T90, fuel
          octane,  oxygen content,  paraffins, olefins, RVP and
          aromatics on emissions.

     Additional information on the scope and intent of these
studies can be found in Air Docket A-92-12  [#II-G-1, December
1990; #II-G-2, January 1992; #II-G-3, January 1992; #II-G-4,
February 1992; #II-G-5, April 1992; #II-G-6, July 1991;  #II-G-7,
July 1991; #II-G-8, #II-G-9, February 1992; #II-G-10, February
1991; and #II-G-11, February 1992]  .


b.   Excluded Studies

     Not all of the studies received by EPA were used to develop
the Complex Model.  Studies were excluded from the database if
they did not meet the following conditions:

          The study must have based its emissions measurements on
          appropriate Federal Test Procedure  (FTP) driving
          cycles,  vehicle preparation cycles, and measurement
          procedures.  Emission measurements on different driving
          cycles are not comparable to one another.

          The study must have tested at least one 1990 technology
          vehicle  [see EPA memorandum "1990 Baseline Vehicles,"
          by Korotney and Machiele in Air Docket A-92-12,
          November 30, 1993], since vehicles that did not have
          1990-equivalent technology were excluded from the
          database.  As discussed in the preamble to this rule,
          Congress specified that the emissions performance of
          fuels were to be based on their performance in 1990
          vehicles.  EPA has interpreted this requirement to
          extend to vehicles from other model years having
          emission characteristics and emission control
          technologies that were present in 1990 model year
          vehicles, since such vehicles should respond similarly
          to fuel modifications.

          The study results must have been submitted to EPA in
          time for development of the Complex Model, i.e., by
          June 1993.
                               105

-------
          The study must have included vehicles from more than
          one manufacturer in order to avoid bias that would not
          be captured in EPA's technology group distinctions
          (discussed in Sections IV.A.2.d and III.B.4).  For
          example, each manufacturer employs different catalyst
          designs and computer controls that are not reflected in
          EPA's technology group distinctions.  A study which
          included only one manufacturer's vehicles might detect
          emission effects peculiar to that manufacturer's
          vehicles.

          The study must have included sufficient quality control
          measures to separate fuel effects from vehicle effects.
          Vehicle effects account for as much as 90% of the
          observed variation in emissions and could, if not
          properly controlled, overwhelm the effects of fuel
          modifications.

Studies received by EPA that failed to meet one or more of these
criteria included: (1)  The NIPER study, which contained no
vehicles with 1990-equivalent technology, (2)  Toyota's
reformulated gasoline study, which included a small number of
Toyota vehicles and no vehicles from other manufacturers, and  (3)
Auto/Oil's high emitter study, which did not maintain adequate
quality controls as discussed in Section A.I.e.


c.   Excluded Tests

     As discussed previously, all data were included in the
database unless strong evidence of inaccuracy, gross error, or
unrepresentative data existed.  EPA recently revised its
definition of volatile organic compounds (VOCs) [see 57-FR-3941,
February 3,  1992 for details] to exclude methane,  ethane and
various other substances, since these compounds have reactivities
that are considerably less than other hydrocarbon compounds.
Since the Clean Air Act requirements for reformulated gasolines
call for reductions in "ozone forming volatile organic compounds"
(hereafter VOC's), EPA also excluded data in specific tests from
the database used to model VOC emissions if such tests did not
measure methane as well as total hydrocarbons.  EPA has found
that the methane fraction of total hydrocarbon emissions tends to
be large and displays considerable variation across different
fuel reformulations.   For tests that measured methane and total
hydrocarbons, the methane fraction averaged 14% but ranged from
6% to 22%.  EPA has been unable to develop a model that
accurately predicts methane emissions or methane fraction of
total hydrocarbon emissions.  Ethane, by contrast, comprises a
relatively small portion of hydrocarbon emissions, averaging
2.95% by mass of nonmethane hydrocarbons.  In addition, EPA has
found the ethane fraction to be relatively stable (±0.9%) across
different fuel reformulations.

                               106

-------
     Because of this variation in methane fraction, EPA has
concluded that a model based on total hydrocarbons cannot be used
to estimate VOC (i.e., nonmethane, nonethane hydrocarbons).  EPA
has concluded, however, that a model based on NMHC can be used to
estimate VOC since ethane is a small and relatively constant
fraction of NMHC.   Some studies measured NMHC emissions directly;
other studies measured total hydrocarbon and methane emissions
separately, which allows calculation of NMHC via simple
subtraction.  Approximately 20% of the data available to EPA in
studies that met the requirements outlined in Section A.I above
contained neither separate methane and total hydrocarbon
measurements nor direct NMHC measurements.  These data were
excluded from the database used to develop the complex exhaust
VOC model.

     In certain situations,  the equivalence ratio  (EQR),  a
calculated quantity which measures the ratio of the actual
air:fuel ratio to the ideal air:fuel ratio, has been found to
have a very strong influence on emissions  [A-92-12 Docket
submittals #IV-E-3,  December 1992; #II-E-6, December 1992; and
EPA memorandum by Rao and Sklar "Analysis of EQR for high
emitting vehicles,"  December 1993 ].  High variability in EQR is
correlated strongly with high variability in measured emissions.
High variability in EQR occurs more often in higher emitting
vehicles than in normal emitting vehicles  (the distinction
between normal and higher emitters is discussed in greater detail
in Sections IV.A.2.d and III.B.4).  EPA's analysis showed that
higher emitting vehicles that are not stabilized  [A-92-12 Docket
submittals #II-B-14, February 1993;  and #II-B-12, February 1993]
prior to emissions testing exhibit large variations in both EQR
and exhaust emission levels.  For this reason, data from higher
emitting vehicles which were not stabilized prior to being tested
were excluded from the database.  This led to the exclusion of
the entire Auto/Oil  high emitter study, as well as all data from
higher emitting vehicles tested in the EPA/EF emissions factor
testing program.

     Finally, all data from tests on Fuel W in EPA's ATL-II
program were excluded from the database.  Fuel W was created by
blending pure chemicals, instead of being composed of typical
refinery streams;  over one-fourth of the fuel consisted of pure
cyclohexane.  Since  Fuel W's emissions effects differed
dramatically from the emissions effects from gasolines with
nearly identical sulfur, aromatics,  olefin, E200, E300, oxygen,
and RVP levels, and since Fuel W's chemical composition differed
dramatically from such gasolines, EPA chose to exclude all Fuel W
data from the Complex Model database.


d.   Excluded Vehicles and 1990 Model Year Technology Vehicles
                               107

-------
     Section 211(k)(10)(A) of the Clean Air Act requires that the
effectiveness of reformulated gasolines be determined in
reference to representative model year 1990 vehicles.  As
discussed in the SNPRM [57 FR 13416  (April 16, 1992)], EPA
determined it was reasonable and necessary to look at '1990
technology type' vehicles.  This decision was made because:  (1)
vehicles with similar engine designs, fuel distribution systems,
and emission control systems would be expected to respond
similarly to fuel changes and (2) much more data is available on
emission effects of fuel changes from this expanded set of
vehicles than on 1990 model year vehicles alone [see EPA
memorandum in docket from Lindhjem to Rykowski,  "Effect of
Oxygenates on Emissions," December, 1993].

     Vehicles included in the studies discussed above were
classified as having '1990 technology' if they met both of the
following criteria:

     (1)  The vehicles must be of model year 1986 or later with
          adaptive learning technology, since over 99% of 1990
          vehicles utilized such technology.

     (2)  The manufacturer of the vehicle must have informed EPA
          that one or more of its 1990 model year vehicles had
          engine,  fuel distribution, and emission control system
          designs similar to those on the vehicle in question
          [see EPA memorandum "1990 Baseline Vehicles," by
          Korotney, November 30, 1993, in Air Docket A-92-12 for
          more details].

A list of the included vehicles are given in the vehicles
database portion of the Complex Model database,  which is
reproduced in Appendix IV-F.

2.   General Model Specifications

a.   Dependent Variables

     Based on input from industry, other government agencies and
other interested parties, EPA decided to use the natural
logarithm (In) of gross emissions as the dependent variable for
all pollutants.  Modeling in log space (in which the dependent
variable is the natural logarithm of mass emissions per mile) has
several distinct advantages [see A-92-12 Docket submittals #II-E-
1, August 1992; II-E-2, August 1992; II-E-3, August 1992; II-E-4,
August 1992; and II-E-5,  June 1992].  First, it increases the
explanatory power of the model by increasing the correlation
coefficient between emissions and fuel parameters.  Second,  EPA
found that models in log space have more normally distributed
residuals (the variation in the dependent variable that cannot be
explained by the model) than did models which were not
logarithmically transformed.  A normal distribution of residuals

                               108

-------
allows the use of a wide array of powerful statistical tools and
standard regression techniques, which enhances the quality and
usefulness of the model. Finally, heteroskedasticity  (in which
the variance of the error differs for different values of the
independent variables)  is reduced when emissions are transformed
using natural logarithms.  NMHC and NOx were modeled on a
In(grams/mile) basis while all air toxics were modeled on a
In(milligrams/mile) basis.  As discussed in Section A.I, all
tests for which methane measurements were not available had to be
excluded from the regressions when modeling NMHC emissions.  NMHC
data were converted to VOC by subtracting the average fraction of
ethane.
b.   Independent Vehicle Variables

     Only vehicles with 1990-equivalent technology were used in
developing the regression models for all pollutants.  The
definition of 1990 technology was specified by the each
manufacturer as detailed in Section A.I.  Data from any vehicle
that did not meet the appropriate 1990 technology definition were
excluded from the regressions.

     As discussed earlier,  the vehicle itself is the dominant
influence on emissions from any vehicle-fuel combination.  This
effect was modeled by including separate independent variables
for each vehicle.  Dummy variables were used to account for these
effects, since they enable the isolation of the vehicles' effects
on emissions from fuel effects and are a standard regression
technique for handling categorical variables [see Hocking,  R.R.
(1976),  "The Analysis and Selection of Variables in Linear
Regression," Biometrics, 32,  1-50.], which take on discrete
values.   Some vehicles were tested in more than one study.   In
such cases,  separate dummy variables were included for each
vehicle in the study in order to minimize "drift" in vehicle
emission levels over time and lab-to-lab measurement variation.
For each study-vehicle combination,  a dummy variable was defined
as "1" for observations providing data on the study-vehicle
combination and "0" for all other observations.  In addition, EPA
chose to use a  no-intercept model in order to isolate all
vehicle effects.  EPA found that models with intercepts tended to
be biased and did not properly account for vehicle effects.
Hence the number of dummy variables in the model for each
pollutant equals the number of vehicle-study combinations for
which emissions data on that pollutant were available.


c.   Independent Fuel Variables

     The independent fuel variables were chosen after careful
consideration of data availability,  expected contribution to
emissions based on outside comments and engineering intuition,

                               109

-------
and preliminary test results.  As detailed in Section A.I.a, the
independent fuel variables for exhaust VOC, NOx, and 1,3-
butadiene emissions included oxygen, sulfur,  aromatics, olefins,
E200, E300, and RVP.   For exhaust benzene emissions, fuel benzene
content was added to the list of independent fuel variables.  For
exhaust formaldehyde and acetaldehyde emissions, the independent
fuel variables included  aromatics, olefins,  E200,  E300, sulfur
and RVP, as well as separate variables for oxygen content in the
form of tertiary amyl methyl ether  (TAME),  ethanol  (ETOH), methyl
tertiary butyl ether (MTBE),  and ethyl tertiary butyl ether
(ETBE) .

     The fuel components of exhaust emissions were separated into
two main categories for the VOC and NOx models.  The first
category consisted of the effects of individual fuel parameters
on emissions.  This effect could not always be modeled with a
simple linear term relating the fuel parameter to the natural
logarithm of mass emissions.   In many cases,  an additional
squared fuel term was required to adequately describe the
emissions effect of an individual fuel parameter.  The second
category of fuel terms consisted of interactive effects between
two different fuel parameters.  These terms were modeled as a
product of the two fuel parameters of interest.  Due to the
paucity of toxics data, the poorer relative precision of toxics
emission measurements when compared to NMHC and NOx measurements,
and the limited understanding of fuel effects on toxic emissions,
toxics emission models included only linear fuel terms.
Initially, EPA chose to pre-exclude all interactive terms from
the four toxics models but to allow all squared terms.  However,
EPA analyses  [see EPA memorandum "Alternative Exhaust Benzene
Complex Models," from Rao and Sklar to Air Docket A-92-12,
December 7, 1993]  showed that adding the squared terms to the
toxics models did not improve their explanatory power while
introducing unexplainable behavior into the models.  For these
reasons, the toxics models included only linear fuel terms.  The
specifics of how particular terms were introduced into the model
is discussed in Sections A.3 and A.4 below.

     The fuel variables were modeled in centered form during the
development of the Complex Model (see model development sections
below).   A centered variable is one in which the variable is
expressed as a deviation from some intermediate value in the
regression equations rather than being expressed directly.  EPA
used centered independent fuel variables in developing the
Complex Model to minimize collinearity problems.  Once the
regression were completed, however, EPA chose to simplify the
resulting emission models by removing the centering values.  This
"uncentering" procedure requires algebraic manipulation of the
centered regression equations to remove the centering values,
yielding equations that contain the uncentered variables.  This
process is discussed in Section A.8.
                               110

-------
d.   Emitter Class and Technology Group Distinctions

     Emission studies conducted by EPA and others have shown that
different vehicles have different emission characteristics.
Differences in vehicles' emission control technologies and
differences in the state of operation of those technologies
account for some of these differences.  For these reasons, EPA
hypothesized that fuel modifications would have different
emission effects on different types of vehicles.  To account for
these differences, EPA's February 1993 proposal divided vehicles
into two "emitter classes" (normal and higher emitters) based on
their exhaust emission levels and modeled the response of each
emitter class to fuel modifications separately.  For the final
Complex Model, EPA modeled higher emitters separately and then
incorporated the adjustment terms developed from the higher
emitter model into an overall in-use model, as discussed in
Section A.3 in greater detail [see A-92-12 Docket submittals #IV-
B-01, August 1992; IV-B-4, July 1993; IV-E-1, June 1992; and IV-
E-6, May 1993 for more details].   These changes were made in
response to EPA's evaluation of its proposed model and comments
received regarding the regression approach used to develop the
NPRM model.  EPA continues to believe that the distinction
between normal and higher emitters to be both valid and necessary
since normal and higher emitters appear to respond differently to
many fuel modifications.

     In the NPRM, EPA further subdivided vehicles in each emitter
class into "technology groups" based on emission control
technology with which each vehicle was equipped.  Categorization
of vehicles into technology groups was used to explicitly account
for the effects of differences in emission control technology,
which were the basis for the technology group definitions.  These
control technologies included catalyst type,  fuel distribution
system, EGR, air injection.  The technology group distinctions
also implicitly accounted for other vehicle characteristics such
as engine design and engine control system, to the extent such
characteristics were correlated with specific technology groups.

     However, as discussed in the February NPRM, EPA was
concerned that technology group distinctions among higher
emitters might not be appropriate, since such vehicles' higher
exhaust VOC emission levels  suggest that their emission control
systems do not function properly.  In addition, the limited
quantity of higher emitter data made it difficult to identify
statistically significant differences in emissions response
between different technology groups within the higher emitter
class of vehicles. The limited higher emitter data caused higher
emitter models developed for individual technology groups to have
low statistical power and poor fit.   Many commenters expressed
identical concerns.
                               Ill

-------
     Efforts to subdivide higher emitters by characteristics such
as exhaust hydrocarbon to NOx ratio, as suggested by several
commenters, did not improve the quality of EPA's higher emitter
model [see EPA memorandum "Proposed Revisions to Technology Group
Definitions for Use in Exhaust Complex Model," from Sklar,
Lindhjem and Rao,  December 2, 1993].  Thus, EPA's final exhaust
models do not divide higher emitters into technology groups.
Technology group specific terms were retained when modeling
normal emitters, for the reasons discussed in the February 1993
Draft RIA and the April 1992 proposal.  However, EPA's final
model limits the use of technology group specific effects to
represent statistically significant deviations from the average
effect across all vehicles,  as discussed more fully in the model
development section (Section A.4).   In response to other
comments, EPA attempted to reduce the number of normal emitter
technology groups. However EPA was unable to identify an
appropriate basis for consolidation and/or combination of the
various technology groups [see EPA memorandum "Proposed Revisions
to Technology Group Definitions for Use in Exhaust Complex
Model," from Sklar, Lindhjem and Rao, December 2, 1993].   EPA
considers its retention of emitter class and technology group
distinctions to be validated by the presence of emitter class and
technology group-specific fuel effects in its final Complex
Models presented below.


e.   Technology Group and Emitter Class Definitions and
     Weightings

     EPA analyzed emissions data on different technology groups
separately since emissions control system designs were expected
to respond differently to changes in fuel parameters. These
differences have been shown to exist in previous EPA analyses
[see EPA memorandum "Effect of Oxygenates on Emissions,"  from
Lindhjem to Rykowski,  December, 1993] as well as in the emissions
predictions from the February NPRM models. Technology groupings
were based on four criteria:

          Fuel delivery system (carburetted, throttle body
          injection or port fuel injection)
          The presence or absence of exhaust gas recirculation
           (EGR)
          The presence or absence of air injection
          Catalytic converter type  (three-way, three-way plus
          oxidation catalyst, etc.)

Of the possible combinations based on these four criteria, only
the nine most common technology groups were used to develop the
model.  These nine technology groups encompassed over 99% of 1990
model year sales of cars and light-duty trucks  (as discussed in
Section E and in the April 1992 SNPRM [92] ) .  Table IV-1
                               112

-------
summarizes  the  definitions of these  nine technology groups  and
the percentage  of 1990 vehicle sales accounted for by each  group.

                             Table IV-1
          Technology Group Definitions and Sales Weights"
TECHNOLOGY
GROUP
1
2
3
4
5
6
7
8
9
FUEL
SYSTEM
Multi
(PFI)
Multi
(PFI)
TBI
Multi
(PFI)
Multi
(PFI)
TBI
TBI
TBI
Garb.
CATALYST
3W
3W
3W
3W + OX
3W
3W
3W + OX
3W
3W +OX
AIR
INJECTION
No Air
No Air
No Air
Air
Air
Air
Air
No Air
Air
EGR
EGR
No EGR
EGR
EGR
EGR
EGR
EGR
No EGR
EGR
PERCENT
1990
SALES
26.31
28.71
21.81
11.43
4.31
0.53
5.25
0.00**
1.65
     Sales weights based on a non-California, 49 state analysis.  See EPA
     memorandum  Number IV-B-2 in Docket A-92-12 for more details.
     Projected sales for technology group 8 were non-zero, but actual reported
     1990 sales for this technology group were zero.
where
     Multi   =  Multi-Port Fuel Injection
     TBI     =  Throttle Body Injection
     Garb.   =  Carburetted

     3W      =  Three Way Catalyst
     3W+OX   =  Three Way Catalyst plus  an Oxidation Catalyst

     Air     =  Supplementary air injection present
     No Air  =  Supplementary air injection not present

     EGR     =  Exhaust gas recirculation used
     No EGR  =  Exhaust gas recirculation not used
     EPA has  also analyzed the effects  of VOC emissions levels  on
vehicle responses to fuel modifications.   Higher vehicle
emissions  indicate that certain control technologies may not  be
functioning properly.   Such vehicles  have been shown in previous
                                113

-------
EPA analyses [see EPA memorandum by Lindhjem "Effect of
Oxygenates on Emissions," December, 1993]  and in the February
NPRM models to exhibit different emission responses to changes in
fuel properties than normal emitting vehicles equipped with
similar emission control technology.  Depending upon the failure
mode causing a vehicle to be a higher emitter,  higher emitters
may also exhibit different emission responses to changes in fuel
properties than other high emitting vehicles.  Because reliable
data was not available as to the type of failure mode for
vehicles considered to be higher emitters,  vehicles were
classified as either 'normal'  or 'higher'  emitters based on their
total hydrocarbon emissions on a CAA-type base fuel on a standard
FTP test.  Vehicles with total hydrocarbon emissions less than
twice the applicable standard (0.82 g/mi)  were classified as
normal emitters.  EPA modeled normal and higher emitting vehicles
separately to account for differences in the responses of such
vehicles to fuel modifications.

     As discussed in the February 1993 proposal, the weight
assigned to each technology group or emitter class for modeling
purposes was set equal to its contribution to in-use emissions.
The weighting factor assigned to normal emitters was then broken
down further by technology group, again according to their
projected contribution to in-use emissions.  These estimates are
very similar to the numbers shown in the February 1993 proposal,
although small changes have been made to account for more
complete information about the fraction of 1990 sales accounted
for by each technology group.   The final in-use eighting factors
for the eight normal emitter groups and the one higher emitter
group are shown below in Table IV-2.  The rationale, derivation
and normalization of the weighting factors are discussed in more
detail in Section III.B.4.
                               114

-------
                            Table IV-2
      Vehicle Group Weightings for the Exhaust Complex Model
Vehicle
Group
Norm*-tg** 1
Norm-tg 2
Norm-tg 3
Norm-tg 4
Norm-tg 5
Norm-tg 6
Norm-tg 7
Norm-tg 9
Higher
emitters
Phase I
VOC/Toxics
0.13986
0.12517
0.09592
0.09508
0.01475
0.00203
0.03905
0.00814
0.48000
NOx
0.19550
0.23430
0.14958
0.15023
0.02400
0.00349
0.04608
0.01682
0.18000
Phase II
VOC/Toxics
0.11942
0.10688
0.08190
0.08118
0.01260
0.00173
0.03334
0.00695
0.55600
NOx
0.17595
0.21087
0.13462
0.13521
0.02160
0.00314
0.04147
0.01514
0.26200
     " designates normal emitters
 "tg" designates technology groups

3.    Development Method Overview

     The overall process used by EPA to arrive  at  the  final
Complex Models for VOC, NOx and air toxics  is summarized  below.
A more detailed discussion of the process can be found in the
following sections.

1.    A "raw" model was developed which contained all regression
     terms found to be significant in  (1) a forward  "stepwise"
     fit, (2) a re-regression on surviving  terms after data
     outliers and influential points were excluded,  (3) a re-
     regression after deleting terms based  on Mallows'  Cp
     criterion to balance overfitting and underfitting, and  (4)  a
     final backwards fit to eliminate insignificant  terms.   Steps
     2-4 were not taken during development  of the  February 1993
     NPRM model.  These steps are discussed in  detail  in  Section
     A.4.

2.    The raw model was refined by deleting  terms whose
     contribution to the model's explanatory power was small.   In
     general, terms which contributed less  than one  percent  of
     the model's explanatory power were deleted to produce a
     refined model which was simpler in form than  the  raw model
     without sacrificing its ability to predict the  emissions
     impact of fuel modifications.  Note that this step was  not
                               115

-------
     taken during development of the February 1993 NPRM model.
     These steps are discussed in greater detail in Section A.5.

3.   The refined model was simplified.  Several commenters
     expressed concern about the complexity and number of
     equations in the final Complex Model.  To address these
     concerns, two steps were taken.   First, the technology
     group-specific models for normal emitters were consolidated
     into a single model for all normal emitters.  Second, the
     fuel variables were uncentered.  Both consolidation and
     uncentering are discussed in Section A.8.

The resulting exhaust emission models contain two equations for
each pollutant (one for normal emitters and one for higher
emitters).   By contrast, the model proposed in February 1993
contained as many as sixteen separate equations for each
pollutant.   In addition, the final exhaust emission equations
have fewer fuel terms than did the February equations.  However,
EPA does not believe that this less complicated complex model is
any less accurate than the complex models presented at public
workshops or in the February 1993 proposal.  This belief is based
on the final models' comparable explanatory power and accuracy in
accounting for the emission effects seen in the vehicle testing
programs that comprise the Complex Model database, as discussed
in Section F.

4.   Raw Model Development

     As discussed previously, exhaust emissions are affected by
both vehicle characteristics and fuel characteristics.  This
combined effect causes difficulty in trying to isolate the
effects of fuel characteristics on emissions.  EPA's analysis
attempted to separate the vehicle and fuel components of exhaust
emissions by using dummy variables to isolate the effect of
vehicles on emissions.  The vehicle was found to be the single
most significant determinant of the level of emissions from a
given vehicle on a given fuel, accounting for approximately 90%
of the variation in emissions.  Fuel properties were found to be
a much smaller (5-6%) influence on variation in exhaust
emissions.   The remaining 4-5% of the variation in emissions
remains unexplained.

     EPA received a number of comments regarding the models
proposed in February 1993.  After careful consideration, EPA
concluded that the following comments were valid and should be
incorporated into its final models:

     •    Some commenters noted that many emission effects were
          likely to be consistent across technology groups or
          emitter classes.  They suggested that EPA include such
          common effects in its models before resorting to
          technology group-specific or emitter class-specific

                               116

-------
          terms to model vehicles' emissions response to fuel
          modifications.

     •    Commenters noted that insufficient data were available
          to model many potential terms, particularly interactive
          terms, and that such terms should not be allowed to
          enter the model.

     •    Commenters argued that in the case of higher emitters,
          EPA's models relied on a small fraction of the database
          to model a large fraction of the in-use emissions.
          They argued that EPA should have used the more complete
          data on normal emitters to model higher emittters'
          response to fuel modifications.

     While EPA does not believe that normal emitter data alone
should be used to build the higher emitter models, EPA agrees
that the normal emitter model, which is based on a more extensive
and complete data set than is available for higher emitters,
should form the basis of the higher emitter model.  EPA used data
on higher emitters to develop adjustment terms that reflect the
difference between normal and higher emitter responses to fuel
modifications.

     To incorporate the concerns raised in these comments,  EPA
has utilized a modified version of the "unified" approach
advocated by a number of commenters to develop its final complex
models.  The techniques used to address these comments are
summarized below and are discussed in greater detail in Sections
A.4.a and A.4.b:

1.    Interactive terms were permitted to enter the model only
     when sufficient data were available.  The February 1993
     models included all statistically significant interactive
     terms, regardless of whether sufficient data were available
     to estimate such an effect.

2.    Models for higher emitters were constructed based solely on
     data from such vehicles in order to determine which fuel
     terms had statistically significant effects.

3.    The entire database was analyzed using the "unified"
     approach.   The effects of each term on emissions was divided
     into two parts:  an average effect across all vehicles, and
     a series of adjustment terms for each technology group and
     for higher emitters.  Higher emitter adjustment terms were
     forced into this model for all terms found to be
     statistically significant in the higher emitter model, in
     order to assure that such effects were not obscured by the
     disproportionately large share of the data on normal
     emitters.
                               117

-------
4.   Overly influential data were dropped from the database and
     the model was re-estimated based on the remaining data (see
     Section A.4.a for further discussion of influential data).

5.   Terms were deleted from the developed model to avoid
     overfitting and collinearity problems.  Mallows'  Cp
     criterion was used to assess the problem of overfitting.

6.   The overall fuel terms and the corresponding fuel adjustment
     terms (for the different technology groups) were computed in
     the following manner:

     a.   An overall effect for the fuel term was estimated using
          all available data in the database.

     b.   The difference between the overall effect and the
          effect for each vehicle group was identified.

     c.   If the difference was significant for one or more
          vehicle groups,  the vehicle group adjustment term with
          the difference that is the most significant was chosen.

     d.   Repeat steps b-c for the remaining vehicle groups until
          there are no more significant vehicle group adjustment
          terms.

     These approaches were designed to take advantage of the
benefits of a regression on the entire database (a "fully
unified" approach) without losing sight of the unique behavior of
higher emitters.  Since higher emitters were severely under-
represented in the complex model database compared to their
contribution to in-use emissions, EPA believed that a fully
unified approach would not have represented higher emitter
effects properly.  However, the partially unified approach used
by EPA allowed the Agency to simplify its emission models without
losing explanatory power.   By using technology group-specific
adjustment terms only if the technology group's response to fuel
modifications was sufficiently consistent in nature and different
in magnitude from the average normal emitter response, EPA was
able to rely on the same set of terms to explain fuel effects
across all eight normal emitter technology groups.  In the NPRM
model, different terms were permitted in each technology group-
specific model.  As a result, the model for one technology group
might include one fuel term instead of another fuel term that
might represent the data equally well, while the model for a
second technology group might include the second term but not the
first.  The partially unified approach applied the same term to
both groups and only permitted additional terms to enter the
model if both the common term and the additional terms were
necessary to adequately model the data.  A copy of the SAS
program used to develop the final complex models for all
pollutants can be found in Appendix IV-E.

                               118

-------
a.   Statistical Development of VOC and NOx Exhaust Models

     The final VOC and NOx models are similar to the most
accurate of the three sets of models included in EPA's July 14,
1993 docket submittal [A-92-12 Docket submittal #IV-B-4,  July
1993],  while also including relevant comments regarding specific
aspects of the models [see Section B.10].   The following step-by-
step procedure was used to develop the "raw" VOC and NOx exhaust
emission models.

1.   The database was reviewed to ensure that specific studies
     and observations were excluded according to the criteria
     outlined in Section A.2.  After excluding this data, the
     database contained 5214 observations  for NOx and 4287
     observations for VOC  (see step 3).

2.   Each vehicle in the database was placed into a technology
     group and an emitter class.  Higher emitters were defined as
     those vehicles which emit greater than 0.82 grams/mile of
     total hydrocarbons on a base fuel [see A-92-12 Docket
     submittal IV-B-3, June 1993].  The 0.82 grams/mile
     represents a cutpoint of twice the VOC emissions standard
     for cars.  The small number of light  duty trucks were
     classified as higher emitters using the same criterion.

3.   Total hydrocarbon (THC) emissions were converted to non-
     methane hydrocarbons  (NMHC) by subtracting measured methane
     emissions from total hydrocarbon emissions.  As discussed
     above, this step can only be performed for those tests in
     which exhaust methane was measured;  other tests were
     excluded from the database used to model exhaust VOC
     emissions.

4.   Fuel oxygenate levels  (in volume percent)  were converted to
     fuel oxygen levels (in weight percent).

5.   Mean values for each fuel parameter were calculated across
     all observations in the database.   These values are given
     below:

                    Oxygen:        1.774834 weight percent
                    RVP:            8.611478 psi
                    E200:           46.72577 percent
                    E300:           85.89620 percent
                    Sulfur:        204.5779 ppmW
                    Olefins:       7.318716 volume percent
                    Aromatics:     28.26110 volume percent

     These mean values were used as the centering values for each
     fuel variable in the regression.
                               119

-------
6.    Separate dummy variables were included in the regression
     models for for each vehicle-study occurrence (for the
     reasons discussed in previous EPA submittals to the docket]

7.    A model for higher emitting vehicles was constructed
     according to the following procedure:
     a.
     b.
     d.
     f.
The vehicle-by-study dummy variable terms were included
in the model.

The following terms were pre-excluded from the higher
emitter VOC and NOx models (details of this step are
discussed in step 10 below):
          OXY*E200
          ARO*SUL
          E200*ARO
          E300*RVP
               OXY*ARO
               SUL*OLE
               E200*OLE
               ARO*RVP
OXY*OLE
SUL*RVP
E200*RVP
SUL*E200
E200*E300
OLE*RVP
A stepwise regression on the higher emitter data was
performed, with the first step including all linear
fuel terms.

In subsequent steps, fuel terms which were
statistically significant at the 0.10 level were added
to the model.  The first step included the vehicle-
study dummy variable terms  (from step a)  and the linear
fuel terms.   Next, higher order terms (squared and two-
parameter interactive terms) were added to the model
using a stepwise regression procedure if they were
significant at the 0.10 significance level, in keeping
with standard EPA practice.  The February 1993 proposal
used a significance level of 0.15.

Once the stepwise model for higher emitters had been
          constructed,  Mallows'  Cp  criterion  was  used  to
                                           The Cp criterion  was
                                              [1973),  "Some
eliminate terms from the model.
proposed by Mallows  [Mallows, C.L.
Comments on Cp," Technometries, 15, 661-675] for
selecting a model and is a measure of total squared
error.  If Cp is  plotted against  the  number of
independent parameters  (p)  in the model, Mallows
recommends the model where Cp first approaches  p.   When
such a model is chosen,  the parameter estimates are
most unbiased and the risk of not including an
additional and possibly valid parameter is balanced by
the risk of adding an invalid parameter.

A backwards-elimination regression was run to remove
linear fuel terms that were no longer significant at
the 0.10 level, unless such terms were necessary for
hierarchy considerations (as discussed below).
                               120

-------
     g.    A separate higher emitter adjustment term was forced
          into the "unified" models for VOC and NOx described
          below for each fuel term remaining in the corresponding
          higher emitter model resulting from step f.   The
          adjustment terms were the interactions of each
          significant fuel term with a dummy variable  designating
          the higher emitter vehicle group.

8.    A "fully unified" model was constructed as discussed below
     in steps 9-14.

9.    The vehicle-study terms and the overall linear fuel terms
     were forced into the model.  Adjustment terms for higher
     emitters based on the results of step 7 were also forced
     into the model.

10.   Certain interactive fuel terms were not permitted to enter
     the model,  i.e., they were pre-excluded from the  model.
     Sufficient data did not exist to incorporate all  possible
     fuel terms into the VOC and NOx models.  In addition,
     research into engine-out versus tailpipe emissions indicated
     that some fuel properties affected catalyst performance
     alone,  while other fuel properties affected combustion
     efficiency alone.  Interactive effects between such
     properties were considered improbable since different
     vehicle components are involved.  Statistical considerations
     such as variance inflation also indicated that certain fuel
     terms should not be permitted to play a role in the models.
     Variance inflation is an indication of the degree of multi-
     collinearity between two or more fuel parameters.  "Large"
     values (in excess of ten) of variance inflation could lead
     to serious distortions of regression coefficients.  For
     these reasons, which are discussed in greater detail in
     EPA's July 14, 1993 docket submittal, the following terms
     were pre-excluded from the normal emitter NMHC and NOx
     models:

          OLE*SUL        RVP*SUL        E200*SUL
          ARO*E200       ARO*RVP        OLE*RVP
          RVP*E300       E200*OXY       E200*E300
          OLE*E200       E200*RVP

11.   The stepwise regression procedure in SAS was used to add
     terms to the models.  Non-linear overall fuel terms  (e.g.,
     two-way interactions and squared terms) and normal-emitter
     adjustment terms specific to individual vehicle groups were
     added if they were statistically significant at the 0.10
     level.   The terms described in step 10 were not permitted to
     enter the model, either as overall terms or adjustment
     terms.
                               121

-------
12.   As with the higher emitter model developed in step 7,  the
     model resulting from step 11 was evaluated using Mallows' Cp
     criterion to balance overfitting and underfitting.   This
     step also helped remove terms with significant variance
     inflation values,  which are indicative of collinearity
     between terms in the model.

13.   Outliers and influential points were removed from the data
     set and the coefficients for the terms in the model
     generated in step 12 were re-estimated.  Emissions results
     that are very different from model predictions can have a
     large impact on the resulting model.  These extreme results
     can be caused by a number of factors that should not be
     permitted to affect the Complex Models:

          •    Errors may have been made in measuring or
               recording emission or fuel parameter levels

          •    Vehicles may have been placed in the wrong
               technology group or emitter class,  or it could
               have had pre-1990 emission control technology

          •    The results may be due to the inherent variability
               of vehicle emissions.

     Such errors can occur during tests of fuels that are
     significantly different from other fuels or of vehicle types
     that are relatively uncommon.  In such cases, the erroneous
     results can distort the regression results significantly
     since countervailing data from other fuels or vehicles would
     not be available.

     This problem was addressed by removing such observations
     from the database and re-estimating the regression equations
     with the remaining data.  Outliers were defined as having an
     Rstudent value greater than an absolute value of 4  (i.e., they
     lie more than four standard deviations away from the mean
     regression result).   The probability of a deviation of this
     size being due solely to the inherent variability of vehicle
     emissions is less than 0.5 percent.  Influential points were
     defined as having a DFFITS value greater than 1.2.   DFFITS
     provides an estimate of the degree to which the observation,
     by itself, alters the regression equation.  Standard
     statistical practice is to consider observations with DFFITS
     values in excess of 1.0 to be overly influential.  EPA
     relaxed this criterion slightly in recognition of the
     limited data available for many fuel parameter combinations
     and vehicle types, thereby retaining some observations that
     arguably could have been excluded.  The number of data
     excluded in this step was small relative to the total number
     of data available, amounting to approximately 1.6% of the
     normal emitter VOC data, 1.8% of the higher emitter VOC

                               122

-------
     data, 1.0% of the normal emitter NOx data, and 0.9% of the
     higher emitter NOx data.

14.  To ensure hierarchy (put simply, ensuring hierarchy requires
     that if an interactive fuel term A*B is in the model,  the
     individual fuel parameters A and B also must be included in
     the model) while also ensuring that the model was
     parsimonious in its choice of terms, a series of backwards-
     elimination runs were conducted to eliminate terms in the
     model developed in step 13 that were no longer statistically
     significant at the 0.10 level.  Such terms include the
     linear overall terms and the higher emitter adjustment
     terms,  which were forced into the model initially.  In
     addition, the variance inflation for the remaining terms was
     examined to assure that the models were free of severe
     multi-collinearity problems.

     In the first backward elimination run, the nonsignificant
     nonlinear fuel terms were removed from the model.  Mallows'
     Cp criterion was  used  to determine  the useful  terms.   In the
     second backward elimination run, the nonsignificant higher
     emitter adjustment terms were removed from the model.
     Again,  Mallows' Cp criterion was utilized.   In the final
     backward elimination run, the nonsignificant nonhierarchial
     linear fuel terms  (i.e., those not included in any
     significant higher-order fuel terms) were removed from the
     model.   Mallows'  Cp criterion was utilized to  balance
     overfitting and underfitting.

EPA also reviewed the resulting models for their statistical
soundness and overall reasonableness.  The SAS code used to
arrive at these models can be found in Appendix IV-D-4.  The
regression coefficients resulting from this methodology are shown
in Section A.5 below.
b.   Statistical Development of the Toxic Exhaust Models

     EPA modeled exhaust toxics emissions in a similar fashion to
the method used to develop the VOC and NOx models.  All five of
the toxic air pollutants that the Clean Air Act specifies for
control through reformulated gasoline (benzene, acetaldehyde,
formaldehyde, 1,3-butadiene and polycyclic organic matter) are
also considered VOCs and are present in exhaust emissions.
Benzene is a natural component of gasoline,  so exhaust benzene
emissions include both unburned benzene and benzene formed in the
combustion process through the partial decomposition of more
complex aromatic compounds.  The four other toxics are not
present in gasoline and hence are solely products of combustion.

     The work discussed in the SNPRM [92] indicated that exhaust
toxics were best characterized in terms of the mass fraction of

                               123

-------
total VOC emissions.  This mass fraction could be multiplied by
total VOC emissions to determine mass emissions of toxic
compounds.  During development of the Complex Model,  however,  EPA
and others found that exhaust toxics models based on milligrams
emitted per mile yielded better correlations than models based on
the fraction of total VOC emissions.  For this reason,  EPA chose
to construct models to estimate the natural logarithm of toxics
emissions (expressed in terms of mg/mile).

     Only a few of the studies described in Section A.I measured
exhaust toxic emissions.  Hence less data were available for
toxics emissions from both normal and higher emitters than were
available for VOC or NOx emissions.  As a general rule, less data
for exhaust toxics than for exhaust VOC or NOx meant that fewer
independent fuel variables could be included in the toxics
models.  For its final toxics models, EPA elected to use only
linear fuel terms, eliminating the squared and the two-parameter
interactive terms found in the models with more data.  EPA
concluded that the resulting increase in the accuracy of the
linear term coefficient estimates more than offset any reductions
in accuracy resulting from exclusion of nonlinear terms.  This
conclusion is supported by EPA's efforts to develop a number of
nonlinear models, all of which had poorer explanatory accuracy
than EPA's linear toxics models [see EPA memorandum by Rao and
Sklar "Alternative Exhaust Benzene Complex Models," December 2,
1993] .

     As with the VOC and NOx models, a "partially unified" model
was built for each exhaust toxic pollutant.  Normal emitter
technology group distinctions were retained while all higher
emitting vehicles were placed in a single group.  The general
methodology used to develop the toxic models was essentially the
same as that used in developing the VOC and NOx models and is
outlined below.

1.   Same as step 1 described in Section A.4.a.

2.   Same as step 2 described in Section A.4.a.

3.   Same as step 4 described in Section A.4.a.

4.   The mean fuel parameter values used were identical to those
     used in step 5 in Section A.4.a for the exhaust 1,3-
     butadiene toxic model.  The exhaust aldehyde models were
     oxygenate specific and thus the following values were used
     as the mean values for the different oxygenates (the other
     fuel parameter centering values are identical to those given
     in Section A.4.a step 5).  The mean value of fuel benzene
     content is given below for use in the exhaust benzene model:

                    MTBE:          0.947240 weight percent
                    ETBE:          0.023203 weight percent

                               124

-------
                    TAME:          0.016443 weight percent
                    Ethanol:        0.314352 weight percent
                    Benzene:        1.066682 volume percent


5.    Same as step 6 described in Section A.4.a.

6.    A higher emitter model was constructed in a manner analogous
     to that discussed in step 7 of Section A.4.  The specific
     steps followed are outlined below:

     a.   Dummy variables for each vehicle-study occurrence were
          forced into the model.

     b.   The following terms were permitted to enter the model
          via a stepwise regression if they were statistically
          significant at the 0.10 level (note that in the
          February 1993 NPRM, terms were included if they were
          significant at the 0.15 level):

          For BENZENE:        Oxygen,  Aromatics, Olefins, E200,
                              E300, RVP, Sulfur and Benzene

          For BUTADIENE:      Oxygen,  Aromatics, Olefins, E200,
                              E300, RVP and Sulfur

          For ACETALDEHYDE:   TAME, Ethanol,  MTBE, ETBE,
                              Aromatics, Olefins, E200,  E300,  RVP
                              and Sulfur

          For FORMALDEHYDE:   TAME, Ethanol,  MTBE, ETBE,
                              Aromatics, Olefins, E200,  E300,  RVP
                              and Sulfur

     d.   Terms were eliminated using Mallows' Cp criterion,
          thereby balancing overfitting and underfitting risks.

     e.   Outliers and influential points,  defined using the same
          criteria described in Section A.4.a, were eliminated
          and the model was re-estimated.

     f.   A backward-elimination regression was used to remove
          non-significant fuel terms.

     g.   For each term identified and retained after step f
          above, force a separate higher emitter adjustment term
          into the "partially unified" model described below.

7.    "Unified" models for each toxic pollutant were built, as
     described in steps 9-14.  The process  was similar to that
     used to build the VOC and NOx models,  as described in
     Section A.4.a.  Outliers and influential points that were

                               125

-------
     removed in step 7 when building the high-emitter models
     described in step 7 were re-included in the database.  They
     were removed in step 12 if they were still outliers.

8.   The vehicle terms and the higher emitter adjustment terms
     from step 7 were forced into the models.

9.   Models for each toxic pollutant were built stepwise, in a
     manner similar to the way the VOC and NOx models were built.
     The following terms were allowed to enter the model as
     either overall terms or as adjustment terms for normal
     emitter technology groups via the stepwise process if they
     were statistically significant at the 0.10 level:

     For BENZENE:        Oxygen,  Aromatics,  Olefins, E200, E300,
                         RVP, Sulfur and Benzene

     For BUTADIENE:      Oxygen,  Aromatics,  Olefins, E200, E300,
                         RVP and Sulfur

     For ACETALDEHYDE:   TAME, Ethanol,  MTBE, ETBE, Aromatics,
                         Olefins, E200,  E300, RVP and Sulfur

     For FORMALDEHYDE:   TAME, Ethanol,  MTBE, ETBE, Aromatics,
                         Olefins, E200,  E300, RVP and Sulfur

10.  The Cpcriterion was applied to the models resulting  from
     step 9.   This step resulted in terms being eliminated from
     the models when the models'  risk of being overfitted
     exceeded the risk of underfitting that would result from
     elimination of the term.

11.  The model generated in step 10 was used to determine which
     observations were outliers or influential points, as defined
     in step 13 in Section A.4.a.  These outliers and influential
     points were removed from the database.

12.  A final stepwise backwards-elimination process was pursued
     to remove fuel terms that were no longer statistically
     significant at the 0.10 level until all remaining terms are
     statistically significant.  This model was free of variance
     inflation problems.

13.  Same as step 13 described in Section A.4.a.


5.   Raw Version of the Exhaust Emissions Complex Model

     Following the steps outlined in Section A.4 resulted in the
derivation of the raw regression model coefficients shown below:
                               126

-------
Raw Regression  Coefficients for  Exhaust  VOC Emissions

                               Table IV-3
Fuel
Term
Oxyqen
Sulfur
RVP
E200
E300
Aro
Olef ins
Aro*E300
E2002
RVP*Oxy
Sul*E300
Oxy*Sul
Sul2
Aro2
E3002
Centered Raw Model Reqression Coefficients for Exhaust VOC
TG* 1
- .00293
.000531
.03274
- .00782
- .00757
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
0
0
.000525
TG 2
- .00293
.00035
- .00006
- .00430
- .00757
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
-1.3E-6
0
.000525
TG 3
- .00293
.000531
.03274
- .00430
- .00757
.00316
- .00251
- .00061
.000086
- .00789
.000016
0
0
0
.000525
TG 4
- .00293
.000802
.03274
- .00430
- .00757
- .0008
- .00251
- .00029
.000086
- .00789
.000016
-9.3E-5
0
- .00031
.000525
TG 5
- .00293
.00090
.03274
- .00430
- .00757
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
0
0
.000525
TG 6
- .00293
.000187
.03274
- .00430
- .00757
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
0
0
.000525
TG 7
- .00293
.000531
.03274
- .00430
- .00757
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
0
0
.000525
TG 9
- .00293
.000531
.03274
- .00430
- .00757
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
0
0
.000525
HIGH**
- .00293
- .000014
.07378
- .00430
.00023
.00316
- .00251
- .00029
.000086
- .00789
.000016
0
0
0
.000525
  11 TG" denotes EPA
  "HIGH"  refers to
technology groups
 higher emitter model
coefficients
                                   127

-------
Raw Regression Coefficients for Exhaust NOx Emissions



                            Table IV-4
Fuel
Term
Oxyqen
Sulfur
RVP
E200
E300
Arom
Olef ins
Sulfur2
Arom2
E3002
Oxy*RVP
Oxy*Aro
Oxy*Sul
E300*01e
Oxy2
Olef ins2
Centered Raw Model Reqression Coefficients for Exhaust NOx
TG 1
.00309
.000706
.011857
.000915
.002465
-.00113
.00234
-1.7E-6
-5.8E-5
0
0
0
-9.4E-5
.000335
0
.000417
TG 2
.00309
.00046
.011857
.000915
-.00014
.002143
.00234
-l.OE-6
-.00021
0
0
-.00099
0
0
0
.000417
TG 3
-.01890
.00046
.011857
.000915
.002465
.002143
.00234
0
-5.8E-5
0
0
0
0
0
.003359
.000417
TG 4
.00309
-.00007
.011857
.000915
-.00312
.002143
.00234
0
-5.8E-5
0
0
0
0
0
0
.000417
TG 5
.00309
.00046
.011857
.000915
.002465
.002143
.00234
0
-5.8E-5
0
.024415
0
0
0
0
.000417
TG 6
.00309
-1.5E-5
.011857
.000915
.002465
.002143
.00234
0
-5.8E-5
0
0
0
0
0
0
.000417
TG 7
.00309
.00046
.011857
.000915
.002465
.002143
.00234
0
-5.8E-5
0
0
0
0
0
0
.000417
TG 9
.00309
.00046
.011857
.000915
.002465
.002143
.00234
0
-5.8E-5
0
0
0
0
0
0
.000417
HIGH
-.00887
.00027
-.01627
.000915
-.00421
.002143
.00234
0
-5.8E-5
-.00057
0
0
0
0
0
.000417
Raw Reqression Coefficients for Exhaust Benzene Emissions



                            Table IV-5
Fuel
Term
Oxyqen
Sulfur
RVP
E200
E300
Arom
Olef ins
Benzene
Centered Raw Model Reqression Coefficients for Exhaust Benzene
TG 1
0
.001054
0
-.00948
0
.02588
0
.222318
TG 2
0
.000337
0
0
0
.02588
0
.222318
Tg 3
0
.001187
0
-.00578
0
.02588
0
.222318
TG 4
0
.000337
0
0
0
.02588
0
.222318
TG 5
0
.000337
0
0
0
.04859
0
.222318
TG 6
0
.000337
0
0
0
.02588
0
.222318
TG 7
0
.000337
0
0
0
.02588
0
.222318
TG 9
0
-.00195
0
0
0
.02588
0
.222318
HIGH
-.09605
.000337
0
0
.011251
.01188
0
.222318
                                128

-------
Raw Regression Coefficients for Exhaust 1,3-Butadiene Emissions



                            Table IV-6
Fuel
Term
Oxyqen
Sulfur
RVP
E200
E300
Arom
Olef ins
Centered Raw Model Coefficients for Exhaust 1, 3-Butadiene
TG 1
0
.000506
0
- .00731
- .01678
- .00401
.028238
TG 2
0
0
0
- .00731
- .01678
- .00401
.028238
TG 3
0
.000544
0
- .00731
- .00625
- .00401
.028238
TG 4
0
- .00041
0
- .00731
- .01678
- .00401
.028238
TG 5
0
0
0
- .00731
- .01678
- .00401
.028238
TG 6
0
0
0
.00579
- .01678
- .00401
.028238
TG 7
0
0
0
- .00731
- .01678
- .00401
.028238
TG 9
0
0
0
- .00731
- .01678
- .00401
.028238
HIGH
- .06077
0
0
- .00731
- .00806
- .00401
.028238
Raw Regression Coefficients for Exhaust Formaldehyde Emissions



                            Table IV-7
Fuel
Term
Sulfur
RVP
E200
E300
Arom
Olef ins
MTBE
ETBE
ETOH
TAME
Centered Raw Model Coefficients for Exhaust Formaldehyde
TG 1
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 2
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 3
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 4
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 5
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 6
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 7
0
0
0
- .01023
- .00717
0
.046213
0
0
0
TG 9
0
0
0
- .01023
- .00717
0
.046213
0
0
0
HIGH
0
0
0
- .01023
- .00717
- .03135
.046213
0
0
0
                                129

-------
Raw Regression Coefficients for Exhaust Acetaldehyde Emissions

                            Table IV-8
Fuel
Term
Sulfur
RVP
E200
E300
Arom
Olef ins
MTBE
ETBE
ETOH
TAME
Centered Raw Model Coefficients for Exhaust Acetaldehyde
TG 1
.000263
0
0
- .01216
- .00555
0
- .03646
.316467
.249326
0
TG 2
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
TG 3
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
TG 4
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
TG 5
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
TG 6
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
TG 7
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
TG 9
.000263
0
0
- .01216
- .00555
0
0
.316467
.249326
0
HIGH
.000263
0
0
- .01216
- .00555
0
- .05598
.316467
.249326
0
     Exhaust POM (polycyclic organic matter) performance is
calculated with the exhaust VOC regression coefficients since POM
was assumed to respond to fuel modifications in the same manner as
did VOC.  EPA made this assumption since POM, while difficult to
measure accurately, is itself a VOC.  Since data on unique POM
responses were not available, EPA considers the VOC model to
provide the most reasonable estimate of POM emissions.  To
calculate mass emissions of POM for a target fuel, the ratio of
POM to exhaust VOC emissions for the baseline fuel must be
multiplied by the exhaust VOC emissions of the target fuel.  The
ratio of baseline POM emissions to exhaust VOC emissions is shown
in Section III.B.5.
6.
Raw Model Refinement
     Once the Complex Model regression equations were determined,
the Agency took steps to refine the model by eliminating any terms
that did not significantly affect its behavior.  In this complex
model development step,  the Agency simplified the model's
structure while simultaneously preserving its behavior.  EPA chose
to refine its model for several reasons.  First, EPA recognized
that many of the terms in the raw models had relatively little
effect on emissions.  Second, EPA had received numerous comments
asking that the Complex Model be simplified to facilitate its use.
The refinement process used by EPA involved two steps:

     1)    Rank the terms in each pollutant-specific model
          based on their contribution to the model's
          explanatory power.  This step was accomplished
          through the random balance approach to model
          simplification.
                                130

-------
     2)    Execute a least-squares regression on the Complex
          Model database to develop equations that contain
          all the original terms except those terms that
          cumulatively accounted for only a small  (generally
          less than 1%)  portion of the model's explanatory
          power.

This section presents the theory and methodology behind the random
balance algorithm, the results of applying the random balance
algorithm to the Complex Model, and the least-squares regression
approach.

a.   The Random Balance Algorithm

     Dr. H.T. McAdams of the Advanced Computing Center of Argenta
developed a method for simplifying models resulting from
statistical regressions.  Dr. McAdams based this "random balance
algorithm" on the premise that a regression model may include some
terms that are statistically significant but contribute very
little to the model's explanatory or predictive power.  Such terms
increase the model's complexity and can be eliminated while
retaining the essential behavioral characteristics of the model.
This section provides an overview of the random balance algorithm;
a detailed description of the random balance algorithm and its use
in simplifying the Complex Model can be found in Appendix III-l.

     The random balance algorithm, as used to refine the raw
Complex Model, can be summarized as follows:

1)   Generate 2000 random fuel blends whose fuel parameter values
     are distributed evenly across the valid range of the raw
     model.  The valid range of the model is discussed in Section
     D.2 .

2)   Use the raw Complex Model for exhaust emissions to estimate
     the gram/mile emissions for each fuel blend.  These estimates
     are fleet-average estimates; they represent the weighted
     estimates from the eight normal emitter and one higher
     emitter regression equations using the Phase II weighting
     factors.

3)   Treat the emission estimates from step (2)  as a data set, and
     execute a simple, multi-parameter regression  (least squares
     polynomial fit) on the natural log of emissions.  The
     randomly generated fuels from step (1)  operate as independent
     variables in this step.  The resulting single exponential
     equation has the same behavioral characteristics as the nine-
     equation raw model.

4)   Calculate the contribution to the model's overall sum of
     squares from individual fuel terms.  The sum of squares is a
     measure of the deviation of individual data points from the

                                131

-------
     regression curve.  The sum of squares provides an indicator
     of the model's overall explanatory power; the smaller the sum
     of squares, the greater the model's overall explanatory
     capability.  This step is completed by squaring each of the
     regression coefficients generated in step (3) and dividing by
     an appropriate normalizing factor.  The approximate
     orthogonality of the moment matrix resulting from step (2)
     allows this calculation to be made, as described in more
     detail in Appendix IV-A.

5)   Rank the model terms on the basis of their contribution to
     the model sum of squares.

     The result of this process is a list of all the terms in the
exhaust Complex Models, ranked in order of importance.  Those
terms which are least important contribute the least to the model
sum of squares.  Decisions can then be made as to which terms can
be eliminated from the model without altering its behavior or
explanatory power.


b.   Ranking of Raw Complex Model Terms

     The raw Complex Model contained six exhaust pollutant
regression models that were candidates for refinement via with the
random balance algorithm.  Of these six models,  the raw models for
benzene, formaldehyde, acetaldehyde, and 1,3-butadiene were
strictly linear and contained relatively few fuel terms.  The
Agency determined that the models for these four pollutants were
sufficiently simple to not require refinement.  Therefore, only
the exhaust VOC and NOx models were refined via the random balance
algorithm.

     The random balance algorithm yielded a list of terms for the
raw exhaust VOC and NOx models ranked in order of decreasing
contribution to each model's predictive power.  These lists are
shown in Tables IV-9 and IV-10.  In each table,  the first column
of values gives the term name,  the second column gives the
termwise contribution to the model's predictive capability, and
the third column gives the cumulative contribution to the model's
predictive capability.
                                132

-------
                            Table  IV-9
              Term Ranking for Raw Exhaust VOC Model
Term name
RVP
E200
E300
ARO
SUL
E300*E300
OLE
E300*ARO
E200*E200
OX Y* RVP
SUL*E300
OXY
ARO* ARO
SUL*SUL
OXY* SUL
Term Contribution
(percent)
31.56
27.72
14.69
7.60
5.56
4.96
2.66
1.98
1.59
0.87
0.65
0.08
0.05
0.02
0.01
Cumulative Contribution
(percent)
31.56
59.28
73.97
81.57
87.13
92.09
94.75
96.73
98.32
99.19
99.84
99.92
99.97
99.99
100.00
                           Table IV-10
                  Term Ranking for Raw NOx Model
Term name
SUL
OLE
OLE*OLE
RVP
E200
ARO* ARO
SUL*SUL
E300
ARO
OXY
OXY* ARO
OXY* SUL
E300*OLE
OXY* OXY
OXY* RVP
E300*E300
Term Contribution
(percent)
50.98
28.74
8.60
3.47
2 .77
1.76
1.69
0.68
0.56
0.24
0.16
0.15
0.14
0.03
0.02
0.01
Cumulative Contribution
(percent)
50.98
79.72
88.32
91.79
94.56
96.32
98.01
98.69
99.25
99.49
99.65
99.80
99.94
99.97
99.99
100.00
     Tables IV-9 and IV-10 show that some model terms contribute a
great deal to the explanatory power of the models,  while others
contribute very little.  These latter terms add complexity to the
Complex Model equations without a corresponding gain in
explanatory power.  The Agency began the refinement process by
                                133

-------
examining terms which contributed a combined 1 percent or less to
the corresponding models' sum of squares.  The Agency also sought
to retain all linear terms in the refined VOC and NOx models
regardless of how much they contributed to the model sum of
squares in order to satisfy the provisions agreed to by the Agency
in the Agreement in Principle.

     Applying the 1 percent cutoff to the VOC model would have
eliminated all terms below the OXY*RVP term in Table IV-9.  These
terms contributed 0.81 percent of the raw model's sum of squares.
Hierarchy considerations dictated retention of the linear oxygen
term despite its low explanatory contribution.  A number of
commenters questioned whether the OXY*RVP term should be retained,
since fuel oxygen content and RVP were highly correlated in the
Complex Model database.  In addition, the Agency's modeling
efforts indicated that models with the OXY*RVP term exhibited
unusual and unexplainable behavior.  For these reasons, EPA chose
to add the OXY*RVP term to the list of terms to eliminate.
Therefore, the following terms were marked for elimination from
the raw exhaust VOC model based on the results of the random
balance technique:  OXY*RVP,  SUL*E300, ARO*ARO, SUL*SUL, and
OXY*SUL.

     Applying the 1 percent cutoff to the NOx model would have
eliminated all terms below the E300 term in Table IV-10.  These
terms contributed 0.75 percent of the raw model's sum of squares.
However, hierarchy considerations dictated retention of the first-
order aromatics and oxygen terms.  Thus the following terms were
marked for elimination from the raw NOx model: OXY*ARO, OXY*SUL,
E300*OLE, OXY*OXY, OXY*RVP, and E300*E300.


c.   Re-regression of the Raw Complex Model

     Once the raw model was developed as described in the Sections
A.4 and A.5, it was evaluated using the random balance approach as
described in Section A.6 to determine the relative contribution of
individual model terms to the predictive power of the raw model.
Terms retained after using an approximate cutpoint of 99 percent
were then pooled and a regression was run (using SAS) on those
terms to determine the coefficients shown in Section A.7 below for
each of the exhaust pollutants.  Since the toxics models are
linear and only very few terms are present in each of the models,
the random balance technique was not used to refine the toxics
models.  Thus, the regression coefficients for the exhaust toxics
models are identical to those shown in Section A.5.

7.   Refined VOC and NOx Model Coefficients
                                134

-------
Refined Exhaust VOC Model Coefficients
                            Table  IV-11
Fuel
Term
Oxyqen
Sulfur
RVP
E200
E300
Aro
OLef ins
E2002
E3002
Aro*E300
Centered Refined Reqression Model Coefficients for Exhaust VOC
TG 1
- .00363
.000515
.03516
- .00683
- .00824
.00355
- .00286
.000106
.000408
- .00029
TG 2
- .00363
.000274
.01038
- .00360
- .00824
.00355
- .00286
.000106
.000408
- .00029
TG 3
- .00363
.000515
.03516
- .00360
- .00824
.00355
- .00286
.000106
.000408
- .00062
TG 4
- .00363
.000836
.03516
- .00360
- .00824
- .00242
- .00286
.000106
.000408
- .00029
TG 5
- .00363
.000876
.03516
- .00360
- .00824
.00355
- .00286
.000106
.000408
- .00029
TG 6
- .00363
.000209
.03516
- .00360
- .00824
.00355
- .00286
.000106
.000408
- .00029
TG 7
- .00363
.000515
.03516
- .00360
- .00824
.00355
- .00286
.000106
.000408
- .00029
TG 9
- .00363
.000515
.03516
- .00360
- .00824
.00355
- .00286
.000106
.000408
- .00029
HIGH
- .00363
-5.4E-5
.04330
- .00360
- .00035
.00355
- .00286
.000106
.000408
- .00029
Refined Exhaust NOx Model Coefficients

                            Table  IV-12
Fuel
Term
Oxyqen
Sulfur
RVP
E200
E300
Aro
Olef ins
Sul2
Aro2
Ole2
Centered Refined Reqression Model Coefficients for Exhaust NOx
TG 1
.004409
.000747
.009077
.000931
.002738
- .00139
.002597
-1.6E-6
-8.0E-5
.000366
TG 2
.004409
.000461
.009077
.000931
-7.7E-5
.002578
.002597
-1.1E-6
- .00022
.000366
TG 3
- .00982
.000461
.009077
.000931
.002738
.002578
.002597
0
-8.0E-5
.000366
TG 4
.004409
-7.1E-5
.009077
.000931
- .00302
.002578
.002597
0
-8.0E-5
.000366
TG 5
.004409
.000461
.009077
.000931
.002738
.002578
.002597
0
-8.0E-5
.000366
TG 6
.004409
-1.5E-5
.009077
.000931
.002738
.002578
.002597
0
-8.0E-5
.000366
TG 7
.004409
.000461
.009077
.000931
.002738
.002578
.002597
0
-8.0E-5
.000366
TG 9
.004409
.000461
.009077
.000931
.002738
.002578
.002597
0
-8.0E-5
.000366
HIGH
- .00913
.000252
- .01397
.000931
- .00401
.002578
.002597
0
-8.0E-5
.000366
8.   Simplification of the Refined Complex Model

     After the raw exhaust Complex Model was refined (as discussed
in Section A.6)  to eliminate terms that did not significantly
contribute to the model's predictive power, additional steps were
taken to simplify its structure.  The first step was to reduce the
number of vehicle group equations in each pollutant-specific model
through an equation consolidation algorithm.  The second step was
                                135

-------
to convert the centered forms of the regression equations to an
uncentered form.  Both of these simplification steps are described
below.  EPA's reasons for simplifying the refined Complex Model
were similar to its reasons for refining the raw model.


a.   Vehicle Group Consolidation

     The statistical approach to the development of the exhaust
Complex Model resulted in separate equations for each vehicle
technology type and emitter class.  Since there were eight normal
emitter technology type equations and a single higher emitter
equation,  the raw Complex Model had a total of nine vehicle group
equations for each pollutant.  The EPA determined that all nine
vehicle group equations would not be necessary if the behavior of
the model could be duplicated through simpler or fewer equations.
Therefore, a vehicle group consolidation algorithm was applied to
the refined Complex Model.

     The consolidation methodology proposed in the July 15, 1993
submittal to the docket made use of "pseudo-geometric averaging."
In this methodology, the weighting factors wi  (described in
Section III.B.4) were applied directly to the model term
coefficients within the exponentials.  Although not strictly
mathematically correct, this method allowed the weighted
coefficients to be summed across normal emitter vehicle groups,
and resulted in a single equation that approximated the original
set of eight.  Thus the final form of the proposed Complex
Model(s) in the July package contained two equations, one
representing normal emitters and a second representing higher
emitters.

     In applying pseudo-geometric averaging to the Complex Model,
two factors tended to reduce the accuracy of the consolidated
equations:

•    Pseudo-geometric averaging resulted in an equation for normal
     emitters whose emission performance estimates were offset
     with respect to the refined and raw models' estimates.  In
     other words, the consolidated model had a consistent bias
     with respect to the refined and raw models.

•    The accuracy of pseudo-geometric averaging broke down for
     large values of the independent variables  (fuel parameters).

EPA considered correcting the consistent bias through the
application of a fixed correction factor to the consolidated
equations.  However, EPA found no mechanism to correct the
increasing inaccuracy of the consolidated model as the value of
the fuel parameters increased.  These two sources of inaccuracy
implied that pseudo-geometric averaging was not an appropriate
consolidation methodology,  and EPA chose to discard pseudo-

                                136

-------
geometric averaging for the final rulemaking in favor of a more
accurate and reliable consolidation approach.

     EPA followed the same guidelines for these alternative
consolidation techniques that it applied to the pseudo-geometric
consolidation approach:

     1)    The unextrapolated forms of the regression equations
          were used during the consolidation process instead of
          the extrapolated forms.

     2)    EPA chose to consolidate the models into two equations,
          one for normal emitters and a second for higher
          emitters.  Consolidation to a single equation would have
          required that a separate Complex Model be promulgated
          for each Phase of the reformulated gasoline program
          since the vehicle group weighting factors are different
          in each Phase.  In addition, EPA expected any bias or
          inaccuracy to be more apparent if normal and higher
          emitters were consolidated into a single equation than
          if normal emitters alone were consolidated (as had been
          the case when using the pseudo-geometric consolidation
          method) .

     3)    The form and emission response characteristics of the
          consolidated model were to remain similar to the form
          and behavior of the unconsolidated refined model.

     The random balance algorithm, introduced in Section A.6 as a
means of refining the raw models, can also be used to develop
consolidated regression equations.  Consolidation is a by-product
of using random balance to rank model terms in order of
contribution to the model sum of squares.  For the purposes of
regression equation consolidation, the random balance algorithm
can be summarized by the following three steps:

1)   Generate 2000 random fuel blends distributed evenly across
     the fuels space.  The valid range limits discussed in Section
     D.2 provided the maximum and minimum values for each fuel
     parameter.

2)   For each fuel blend, determine the emissions of VOC,  NOx, and
     toxics in grams/mile for the normal emitter exhaust Complex
     Model.  This step results in fleet average emission estimates
     for normal emitters that represent the emission-weighted
     results from the individual regression equations.

3)   Treat the emission estimates from step  (2) as a data set, and
     execute a simple, multi-parameter regression (using a least
     squares polynomial fit) on the natural log of these emission
     estimates.  The fuel parameters for the 2000 fuels from step
     (1) operate as independent variables in this step.

                                137

-------
Thus the random balance algorithm fits a single equation to the
results of a set of equations which duplicates their behavior in a
simpler form.  Details of the random balance algorithm are
provided in Appendix IV-A.

     The random balance algorithm yields a consolidated equation
that will necessarily be biased in its predictions as compared to
the unconsolidated model on which it is based.  This bias arises
because the step (3) regressions are done in natural logarithm
space.  The least squares fit to the natural logarithm of the
emission estimates calculated in step (2) results in normally-
distributed emission estimates about the polynomial regression
equation.  The exponential of the new regression equation is then
taken to convert the natural logarithm of emissions back to
emissions in grams/mile and place it in the same form as the
unconsolidated model.  Exponentiation causes the distribution of
emission performances calculated in step (2) to become skewed in
one direction rather than remaining evenly distributed.  As a
result, the consolidated model tends to estimate slightly larger
emission benefits than the unconsolidated model.

     The bias introduced into the consolidated Complex Model
through the random balance algorithm is of the same magnitude and
direction as the bias introduced by pseudo-geometric averaging.
However, EPA has determined that the merits of consolidation
outweigh the detriment posed by such bias since it is negligible,
as discussed below.  In addition, EPA carried out an analysis of
the extent of the bias and concluded that no bias correction
through fixed correction factors or other means need be
implemented.  The details of the analysis are given below.

     The analysis of the consolidated Complex Model began with the
creation of 500 randomly generated fuel blends in an identical
manner to step (1)  of the consolidation process.  Then both the
consolidated and unconsolidated models were evaluated for each
fuel, and the difference between the two emission performances
were recorded.  The product of the analysis was a set of
performance differences for each pollutant, one for each of the
500 random blends,  equal to the performance of the consolidated
model for a given pollutant and target fuel minus the performance
of the unconsolidated model for the same pollutant and target
fuel.  If no bias were present, the average performance delta
would be zero.  Based on the form of the exhaust Complex Model
equations, EPA expected the bias to result in negative performance
differences.  The average performance difference for summer, Class
C areas is given in Table IV-13.
                                138

-------
                            Table  IV-13
                Consolidation Performance  Differences
Pollutant
Exhaust benzene
Non- exhaust benzene
1, 3 -Butadiene
Acetaldehyde
Formaldehyde
POM
Total toxics
Exhaust VOC
Non -exhaust VOC
Total VOC
NOx
Phase I
-0.0762
0.0
-0.0565
-0.3157
0.0001
0.0
-0.0620
-0.02
0.0
-0.0073
0.0043
Phase II
-0.0651
0.0
-0.0482
-0.2696
0.0001
0.0
-0.0610
-0.02
0.0
-0.0130
0.0043
     Before drawing conclusions from the values in Table IV-13, a
number of comments should be made.  First, the models for
nonexhaust benzene, nonexhaust VOC, and exhaust POM did not
require consolidation, which is why their performance differences
were zero.  They are included in Table IV-13 because it is the
bias for total toxics, total VOC,  and NOx which determines the
accuracy of the consolidated Complex Model compared to that of the
unconsolidated Complex Model.  Second, the expected negative value
for the performance differences was not exhibited for NOx and
formaldehyde.  However, the positive performance differences are
all very small, and most likely result from two sources:

     1)   A consolidated equation that represents the behavior of
          the unconsolidated model very well.

     2)   Random variation in the calculated emission estimates
          for the random fuel blends.

     The performance differences in Table IV-13 must be examined
in terms of their contribution to total toxics, total VOC, and NOx
because it is these three quantities that are regulated by the
reformulated gasoline program.  Thus the average performance
differences were weighted according to the gram/mile emission
values for Clean Air Act (CAA) baseline fuel to obtain values for
total toxics and total VOC.  After doing so, EPA has determined
that the average performance differences for total toxics, total
VOC, and NOx are negligible for the following reasons:

     •    The measurement uncertainty for the Complex Model
          fuel parameters create emission estimate
                                139

-------
          uncertainty that is larger than the error
          introduced by consolidating normal emitter
          technology groups.

     •    The error introduced by consolidation for VOC and
          toxics is small relative to the overall emission
          reductions required by the program for these
          pollutants.

     •    The error introduced by consolidation for NOx is
          favorable, i.e., on average, the consolidated model
          slightly overestimates the emissions from a given fuel,
          though the error for any given fuel may be unfavorable.
          As a result,  fuels that meet the NOx requirements of the
          program under the consolidated model will, on average,
          meet the NOx requirements under the unconsolidated model
          by a slightly larger margin.

For these reasons, EPA decided that the small average performance
differences did not require the application of fixed correction
factors or other measures in order to assure the integrity of the
reformulated gasoline and anti-dumping programs.


b.   Uncentering

     The statistical approach to the development of the Complex
Model included the use of centered independent variables.   A
centered variable is one in which the variable is represented as a
deviation from some fixed value rather than just as its measured
value.  This fixed value is called the "centering value."   Thus
the centered form of the aromatics variable could be written as

                            ARO  - AROC

where ARO is the amount of aromatics in a given fuel in units of
volume percent, and AROC  is the  centering  value  of  aromatics,  also
in units of volume percent.   Similar centered variables can be
constructed for all independent variables in the Complex Model:

     OXY - OXYC          SUL - SULC           RVP - RVPC
     E200 - E200C        E300 - E300C        OLE - OLEC
     BEN - BENC          MTB - MTBC           ETB - ETBC
     ETH - ETHC          TAM - TAMC

Thus the linear, squared, and interactive terms in the Complex
Model take the following forms:

          Linear:        c1 *  (RVP -  RVPC)
          Squared:       c2 *  (SUL -  SULC)2
          Interactive:    c3 *  (ARO -  AROC) *  (E300  - E300J


                                140

-------
where ci;  c2, and c3 are coefficients determined through
statistical regression techniques.  The centering value of each of
the 12 independent variables is approximately the mean value of
that variable measured across all observations in the Complex
Model database.  The precise mean value was not used as the
centering value for three main reasons:

•    The mean values are different for each pollutant, since not
     all observations measured each pollutant.

•    The centering values were calculated before all of the
     studies included in the Complex Model database were made
     available to EPA.  Once these additional studies were
     included in the database, the mean values and previously-
     calculated centering values diverged.

•    If the mean values were used as the centering values, then
     the centering values would have changed as individual
     observations were excluded based on the outlier and
     influential point criteria discussed in Section A.4.

The centering values used in EPA's regressions are repeated below:
                            Table  IV-14
          Centering Values Used in Exhaust Complex Model
Fuel parameter
Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
MTBE
ETBE
Ethanol
TAME
Centering value
1.774834
204.5779
8.611479
46.72577
85.89620
28.26109
7.318716
1.066683
0.947240
0.023204
0.314352
0.016443
     Centered independent variables were used in the development
of the Complex Model to minimize collinearity problems.  Once the
regressions were completed, EPA chose to remove the centering
values in order to simplify the form of the Complex Model
equations.  Algebraic manipulations of the regression equations
yielded equations that contained only the variables themselves,
rather than centered variables.  This "uncentering" process
resulted in a simplified version of the Complex Model equations.
                                141

-------
     A hypothetical  example can help illustrate the uncentering
algorithm.  The  Complex Model exhaust emission equations take the
following  form:

      (A.I)     Y(t)  = [exp(f (t) )/exp(f (b) )  -  1]  * 100

where "Y(t)M  is  the  percent change in the  exhaust emissions of any
exhaust pollutant  Y,  "t" refers to the  set of target fuel
parameter  values,  and "b" refers to the  set of baseline fuel
parameter  values.  The hypothetical second-order function f can be
written in its centered form as

      (A. 2)     f(A,B)  =  Sl(A  - Ac) + e2(B  - Bc) + e3 (A - Ac)2 +
                          e4(B  - Bc)2 + e5(A - Ac) * (B - Bc)

where A and B represent hypothetical fuel  parameters (independent
variables) , and  the  coefficients e± are constants  that  are
determined through statistical regressions.

     The function  f  can be expanded to yield

      (A. 3)     f(A,B)  =  SlA -  SlAc + e2B - e2Bc +
                          e3A2 - 2e3AAc  +  e3Ac2 + e4B2  - 2e4BBc  +  e4Bc2
+
                          e5AB  - e5ABc - e5AGB + e5AGBc

Grouping similar terms yields

      (A. 4)     f(A,B)  =  A * (Sl -  2e3Ac  -  e5Bc)  +
                          B * (e2 -  2e4Bc  -  e5Ac)  +
                          A2  *  (e3)  +
                          B2  *  (e4)  +
                          AB *  (e5)  +
                          (-e^  - e2Bc  +  e3Ac2 + e4Bc2 + e5AGBc)

At this point every  variable in Equation (A. 4)  is expressed in its
uncentered form.   The coefficients for  the second-order terms
remain the same  while the coefficients  for the first-order terms
are now functions  of the e±  and the centering values.   In
addition,  the function f now contains a  constant.   Equation (A. 4)
can be rewritten more succinctly as follows:

      (A. 5)     f(A,B)  = r^A  +  n2B + I13A2  + r]4B2 + n5AB + So

where the  r\i and So are defined  as  follows

           !li = ®i  -  2e3Ac -  e5Bc
           i!2 = e2  -  2e4Bc -  e5Ac
                                22
          So  =   -e^ -  e2Bc + e3Ac2 + e4Bc2 + e5AGBc

                                 142

-------
The iii and So are constants just as the eL and centering values are
constants.  Since the uncentered  form of the function f in
Equation  (A.5)  is mathematically  identical to the centered form in
Equation  (A.2), the centered  form of  the function can be replaced
with the uncentered form  without  altering the model's behavior.

     The Complex Model can be simplified still further by deleting
constants.  Since the exhaust emission performance of a given fuel
is calculated using a ratio of two exponentials,  constants found
in both exponentials will cancel  one  another when calculating
exhaust emission performance.   The percent change in the exhaust
emissions of a  pollutant  Y can be calculated from Equation (A.I)
as follows  (using the hypothetical exhaust emission function f
from Equation  (A.5) to illustrate the point):

      (A.I)      Y(t) =  [exp(f (t) )/exp(f (b) )  - I]  * 100

Incorporating Equation  (A.5)  into Equation (A.I)  yields
(A. 6) Y(t) =    exp(n:Ah + I],Bh + ri3Ah  + ri4Bh  + ^.-A^ + &)   - 1

                * 100

where At indicates the level of fuel parameter A  in  the  target
fuel, while Ab indicates the level of fuel parameter A in the
baseline fuel.  The exponential of  a sum of  two quantities is
defined as the  product of  the  exponentials of each of those two
quantities.  Thus Equation (A.6)  can be written as follows:

                lexpdijAt + n2Bt + n:At2 + n1Bt2  + n;AtBt) exp (So)
(A. 7) Y(t) =    expd^At, + ii2Bb + ii3Ab2 + ii4Bb2  + ii5AbBb)exp(So)  - I I

                * 100

The exp (So) terms cancel  one  another,  and the equation for Y(t)
simplifies to
                                               AtBtl
      Y(t) =    expdiA, + il2Bb + ri3V + ii4Bb" + il5AbBb)   - 1 J * 100
Thus the constant So, which was  introduced into the function f
through the uncentering algorithm,  can be dropped from the
function f altogether.  The  remaining function,  devoid of the
constant So, then contains only  the  uncentered variables and their
coefficients.  Equation  (A.8) is  mathematically identical to the
centered form implied by Equations  (A.I)  and (A.2),  so the
centered form of f can be replaced  with the uncentered form
without altering the model's behavior in any way.


9.   Final Complex Models

                                143

-------
     The simplification methodology described in Section A.8
produces two complex model equations (one for normal emitters and
one for higher emitters)  for each pollutant.  The coefficients
shown below represent the final,  simplified coefficients for each
pollutant.  The details of how to compute in-use performances and
in-use mass emissions for a target fuel from these coefficients
are discussed more fully in Section C.
Final Exhaust VOC Model
                            Table  IV-IS
Fuel
Term
Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
E2002
E3002
Aromatics*E300
Final Exhaust VOC Model Coefficients
Normal Emitters
-0.003641
0.0005219
0.0289749
-0.014470
-0.068624
0.0323712
-0.002858
0.0001072
0.0004087
-0.0003481
Higher Emitters
-0.003626
-0.000054
0.0432950
-0.013504
-0.062327
0.0282042
-0.002858
0.0001060
0.0004080
-0.000287
                                144

-------
Final Exhaust NOx Model
                            Table IV-16
Fuel
Term
Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Sulfur2
Aromatics2
Olefins2
Final NOx Model Coefficients
Normal Emitters
0.0018571
0.0006921
0.0090744
0.0009310
0.0008460
0.0083632
-0.002774
-6.63 x 10"7
-0.000119
0.0003665
Higher Emitters
-0.00913
0.000252
-0.01397
0.000931
-0.00401
0.007097
-0.00276
0
-7.995 x 10"5
0.0003665
Final Exhaust Benzene Model
                            Table IV-17
Fuel
Term
Oxygen
Sulfur
E200
E300
Aromatics
Benzene
Final Exhaust Benzene Model Coefficients
Normal Emitters
0
0.0006197
-0.003376
0
0.026550
0.222390
Higher Emitters
-0.096047
0.000337
0
0.011251
0.011882
0.222318
                                145

-------
Final Exhaust 1,3-Butadiene Model



                            Table IV-18
Fuel
Term
Oxygen
Sulfur
E200
E300
Aromatic s
Olef ins
Final 1 , 3 -Butadiene Model Coefficients
Normal Emitters
0
0.0001552
-0.007253
-0.014866
-0.004005
0.0282350
Higher Emitters
-0.060771
0
-0.007311
-0.008058
-0.004005
0.0436960
Final Exhaust Formaldehyde Model



                            Table  IV-19
Fuel
Term
E300
Aromatic s
Olef ins
MTBE
Final Formaldeyde Model Coefficients
Normal Emitters
-0.010226
-0.007166
0
0.0462131
Higher Emitters
-0.010226
-0.007166
-0.031352
0.0462131
                                146

-------
Final Exhaust Acetaldehyde Model

                            Table  IV-20
Fuel
Term
Sulfur
RVP
E300
Aromatic s
MTBE
ETBE
Ethanol
Final Acetaldehyde Model Coefficients
Normal Emitters
0.0002631
0.0397860
-0.012172
-0.005525
-0.009594
0.3165800
0.2492500
Higher Emitters
0.0002627
0
-0.012157
-0.005548
-0.055980
0.3164665
0.2493259
The final coefficients for exhaust POM are identical to the final
coefficients shown for exhaust VOC in Table IV-15.

     The aldehyde models are oxygenate specific.  In Tables IV-19
and IV-20 the formaldehyde and acetaldehyde models contain MTBE
(methyl tertiary butyl ether),  ETBE (ethyl tertiary butyl ether),
and ethanol terms.  These terms represent the amount of oxygen (in
terms of weight percent) present in the fuel in the form of the
particular oxygenate in question.  Data on the effects of MTBE,
ETBE, ethanol,  and TAME (tertiary amyl methyl ether) on aldehyde
emissions were available in the Complex Model database.  TAME-
specific terms were not found to be statistically significant in
the regression equations for either aldehyde and thus do not
appear in the final models.

     The following procedure should be used to evaluate the
emission performance of fuels containing oxygenates other than
MTBE, ETBE, ethanol, or TAME.   If the oxygenate is any type of
methyl ether other than TAME,  use the MTBE terms to evaluate its
effects on aldehyde emissions.   If the oxygenate is any type of
ethyl ether, use the ETBE terms to evaluate its effect on aldehyde
emissions.  This distinction is made because the ethyl/methyl
group functionality (in the ether compound) is expected to play
the most important role in the combusion chemistry and kinetics
which produce the aledhydes.   If the oxygenate is an alcohol other
than methanol,  use the ethanol terms to evaluate its effect on
aldehyde emissions.  If the oxygenate is neither an ethyl or
methyl ether (i.e, if it is any higher order ether), use the ETBE
terms to evaluate all toxic performances.  This approach was taken
since, given the lack of data on the effects of higher order
ethers on aldehyde emissions,  EPA considers the performance of
                                147

-------
ETBE (which is the highest-order and heaviest ether for which
aldehyde effects data is available in the database) to be the best
available estimate of the aldehyde effects of higher-order and
heavier ethers.  Fuels containing any other oxygenate (including
methanol) in greater than trace amounts cannot be certified using
the Complex Model unless the Model is augmented with the results
of an appropriately designed and conducted vehicle testing
program, as outlined in Section G.


10.  Response to Comments

     A number of comments have been made on issues related to the
development of the Complex Model.  These comments are discussed
separately below for each subject area.

     Although no commenters had objections to vehicle group
consolidation, the proposed pseudo-geometric averaging approach to
consolidation was considered inappropriate by some.  The two main
objections to this approach were:

     1)    It is mathematically incorrect; and

     2)    It results in a bias in the consolidated model that
          might have required the use of performance margins to
          offset this bias.

The Agency showed in the July 15, 1993 docket submission that
although pseudo-geometric averaging is not mathematically correct,
it does in fact result in a consolidated Complex Model that is
sufficiently accurate.  Regardless, pseudo-geometric averaging has
been replaced by the more reliable and accurate random balance
approach to consolidation, eliminating this concern.  In addition,
EPA's analysis of the final, consolidated Complex Model showed
that no margins or offsets need be applied to the regression
equations or the performance standards when the random balance
approach is used for consolidation.  Those parties that commented
on the random balance consolidation approach put forth as an
alternative consolidation method in the July 15 docket submittal
had no objections to its use.  Details of the random balance
approach and the analysis of any resulting bias are given in
Section A.8.

     Commenters also expressed concern about the number of
equations in the final Complex Model.  Several commenters asserted
that a single consolidated equation could represent the essential
behavior of the Complex Model as well as two or more equations.
The Agency chose not to consolidate down to a single equation for
the following reasons (see Section A.8 for more details):
                                148

-------
     •    Many potential users of the Complex Model expressed a
          desire to see separate effects for normal and higher
          emitting vehicles.

     •    The Agency recognized that further simplification would
          yield diminishing returns as the models became simpler.
          Consolidating into a single equation would require the
          promulgation of two unique Complex Models, since the
          vehicle group weighting factors are different for each
          Phase of the program.  Thus the greater simplicity of a
          single equation would be offset by the need to use
          separate equations for each Phase.

     •    The analysis of the consolidated equations showed that
          no offsets needed to be applied to the model to correct
          for bias introduced through random balance consolidation
          to two equations.  However, the same would not have been
          true if the raw model had been consolidated down to a
          single equation.  The differences between normal and
          higher emitter equations are greater than the
          differences among normal emitter technology groups, and
          these greater differences would increase the size of the
          bias introduced through consolidation.  In other words,
          the bias would be expected to increase and become
          significant in the context of the reformulated gasoline
          and anti-dumping programs if the model were consolidated
          to a single equation for each pollutant.

     •    Consolidation from two equations to one equation does
          not significantly increase the simplicity of the Complex
          Model or decrease the calculation time.  A single
          equation also would not facilitate incorporation of the
          Complex Model into linear programming models or
          significantly simplify the derivation and application of
          extrapolation equations compared to two-equation models.

     Some commenters also stressed that "overestimation of
emission benefits" was an inappropriate descriptor of the bias
introduced through consolidation.  These commenters particularly
disagreed with the use of this phrase in describing the proposal
to implement correction factors or offsets to the consolidated
Complex Model.  As described above, this phrase is provided as a
description of the behavior of the consolidated Complex Model in
comparison to the unconsolidated Complex Model only.  In other
words, consolidation results in Complex Model behavior that is
consistently skewed in one direction with respect to the raw
vehicle group equations.  The Agency did not and does not wish to
imply that the consolidated Complex Model provides unwarranted
emission benefits when estimating real, in-use emissions.  In
addition, the proposal to apply margins or offsets to correct for
this bias was presented simply as a mathematical correction
factor.  As stated above, however, the bias is too small to

                                149

-------
require any such correction factors when using the random balance
method to consolidate the normal emitter technology group
equations into a single equation.

     Another set of comments focused on which of the three EPA
options proposed in July [see A-92-12 Docket submittals #IV-B-4,
July 1993 and #IV-B-5, July 1993] was most accurate while at the
same time fungible and simple to use.  Most of the comments
focused on the following points:

     •    Any models that contain extrema points within the
          allowable range of the model should not be consdiered
          due to the possible blending problems that they may
          cause

     •    Option 2 should be chosen for all pollutants since
          maxima values in option 1 result in unexplainable
          behavior

     •    Option 3 is oversimplifed

     •    Option 1 has "superior" aromatics and E300 responses and
          is "most" technically correct and thus should be chosen
          for all pollutants

     •    The effect of sulfur on VOC and NOx has been
          underestimated by all the EPA options

     •    Olefins should be excluded from the model because
          reducing olefins increases VOC emissions in all options

     •    Simplification of emissions equation structure while
          retaining accuracy is very important

     EPA has addressed most of these comments in arriving at the
final exhaust models shown in Section A.9.  The final model chosen
for VOC was the option 1 model with the OXY*RVP term removed,
since this term caused the model to behave in ways not supported
by the available data and since few well-controlled observations
were available to estimate this term.  The final model chosen for
NOx was the option 1 model.  The toxics models are all linear.
EPA believes that the ARO*E300 interactive term in the exhaust VOC
model is needed to represent the aromatics and E300 effects
accurately.  The olefins effect in the final VOC model shows
emissions to increase as olefins are decreased which is what is
expected from data that exist in the Complex Model database.  EPA
believes the squared terms (Sulfur2,  aromatics2 and olefins2)
present in the NOx model are necessary to accurately predict fuel
effects on emissions.  All extrema values (emission maxima or
minima) have been handled by appropriate extrapolation procedures
(see Section D.3). The predictability analysis in Section F.2
indicates that the final models chosen by EPA for VOC, NOx and

                                150

-------
toxics accurately predict the exhaust emission effects from fuel
modifications.  EPA believes that the final models being
promulgated are sufficiently simple without sacrificing accuracy.
B.    Complex Model for Non-Exhaust Emissions

     Vehicle emissions can be separated into two broad categories:
exhaust emissions, resulting from combustion of the fuel, and
nonexhaust emissions, resulting from fuel evaporation, vapor
displacement, and permeation.  The Agency has developed nonexhaust
models using a different process from that used to develop the
exhaust portion of the Complex Model.   The nonexhaust model
predicts emissions of volatile organic compounds (VOC) and benzene
in warmer "Class B" and cooler "Class C" areas; in the
regulations, preamble, and other supporting documents for the
reformulated gasoline rule, Class B areas are referred to as VOC
Control Region I and Class C areas are referred to as VOC Control
Region 2.  The nonexhaust VOC model was derived from EPA's Highway
Vehicle Emission Factor MOBILE models, while the nonexhaust
benzene model was derived from correlations originally developed
from General Motors'  thermodynamic vapor equilibrium model.  The
behavior of the non-exhaust benzene model has been confirmed by
EPA through independent emissions testing.  The remainder of this
section describes the derivation and use of the Complex Model for
nonexhaust emissions.

     EPA subdivides nonexhaust VOC emissions into evaporative,
running loss, and refueling VOC emissions.  These emissions are
primarily a function of fuel volatility (RVP).   Nonexhaust benzene
emissions are the only nonexhaust toxic emissions source regulated
by the reformulated gasoline program and are primarily a function
of RVP, benzene, and MTBE.   MTBE is the only oxygenate currently
known to inhibit benzene vapor pressure at a given RVP level.  The
simple model addressed these very same factors, so the Complex
Model for nonexhaust emissions follows the same development
pattern as that for the simple model:   the Complex Model for
nonexhaust emissions includes only the effects of fuel benzene,
RVP,  and MTBE levels on nonexhaust benzene emissions, and RVP on
nonexhaust VOC emissions.  The July 11, 1991 MOBILE4.1 model with
basic I/M and the MOBILE5a model with enhanced I/M were used to
estimate nonexhaust VOC emissions for the Phase I and Phase II
timeframes,  respectively.  These two MOBILE models have been
described in Section III. The derivation of the Phase I Complex
Model for nonexhaust VOC emissions is described in Section B.I
below, while the analogous Complex Model for Phase II nonexhaust
VOC emissions is described in Section B.2.  The Complex Model for
nonexhaust toxics emissions is covered in Section B.3.

1.    Phase I Nonexhaust VOC Emissions Model
                                151

-------
     The Phase I relationship between nonexhaust VOC emissions and
RVP was developed by fitting a curve to nonexhaust VOC emission
predictions provided by MOBILE4.1.  Since the same technique was
used to develop the nonexhaust portion of the simple model, the
nonexhaust VOC emissions equations in the Phase I Complex Model
are exactly the same as the the nonexhaust VOC equations in the
simple model.

     To assess the relationship between the Reid vapor pressure of
a given fuel and the nonexhaust VOC emissions that such a fuel
would be expected to produce, MOBILE4.1 was run for the year 2015
with summer temperatures and the basic I&M scenario described
previously.  Fuel RVP levels ranged from 6.5 to 8.7 psi (the input
file is shown in Appendix IV).   The year 2015 was chosen as the
target year to allow MOBILE4.1 to model a hypothetical in-use
fleet composed entirely of vehicles with 1990 technology.
Separate results for Class C and Class B areas were produced,
since these areas' different average minimum and maximum summer
temperatures for high ozone days alter the relationship between
RVP and nonexhaust VOC emissions.

     Each set of MOBILE4.1 results gave emission factors in terms
of total grams of VOCs per mile, categorized by vehicle type (LDV,
HDV, etc.) and emission type (refueling losses, evaporative
losses, and running losses) .   Both the simple model and Phase I
Complex Model are based on results for light-duty vehicles  (i.e.,
cars).  Table IV-21 shows the MOBILE4.1 results that were used to
develop the Phase I Complex Model for nonexhaust VOC emissions.
                                152

-------
                            Table IV-21
       Nonexhaust Emissions from MOBILE4.1  (July 11, 1991]


Class

C








B










RVP

8.7
8.5
8.3
8.1
7.9
7.7
7.5
7.2
7.0
8.7
8.5
7.8
7.6
7.4
7.2
7.0
6.8
6.6
VOC emissions for LDGVs in grams
per mile:
Refueling
Losses
0.04
0.04
0.04
0.04
0.04
0.03
0.03
0.03
0.03
0.04
0.04
0.03
0.03
0.03
0.03
0.03
0.03
0.03
Running
Losses
0.39
0.36
0.33
0.31
0.28
0.25
0.23
0.20

0.43
0.40
0.29
0.26
0.23




Evaporative
Losses
0.34
0.31
0.29
0.27
0.25
0.23
0.21
0.19
0.18
0.39
0.36
0.27
0.25
0.23
0.21
0.20
0.18
0.17
     A numeric polynomial regression was carried out for each of
the three emission types (running, refueling, and evaporative
losses)  to produce a relationship giving VOC emissions in grams
per mile as a function of the Reid vapor pressure in pounds per
square inch.  This regression resulted in separate equations for
Class B and Class C areas,  as shown in Table IV-22.
                                153

-------
                            Table  IV-22
   Phase I Nonexhaust Emission Regressions as a Function of RVP
Area
Class
B
B
B
C
C
C
Vapor source
Refueling
Running
Evaporative
Refueling
Running
Evaporative
Curve Fit
0.006668 * RVP - 0.018
0.002791 * RVP2 + 0.1096 * RVP -
0.734
0.02293 * RVP2 - 0.2461 * RVP +
0.7952
0.006668 * RVP - 0.018
0.016255 * RVP2 - 0.1306 * RVP +
0.2963
0.021239 * RVP2 - 0.2393 * RVP +
0.813
R2
0.685
1.000
0.999
0.685
0.998
0.999
     Evaporative emissions within the MOBILE emissions model
include both hot-soak and diurnal emissions, but they are not
given in units of grams per mile.  This conversion requires
calculation of a hot soak/diurnal ratio from the MOBILE4.1
database.  The average weighting over all vehicle ages is 67.9
percent hot soak and 32.1 percent diurnal.  Based on these
proportions, the evaporative emission equations can be split into
separate equations for hot soak and diurnal emissions, as shown in
Table IV-23.

                            Table IV-23
    Evaporative  Emissions Breakdowns  into Hot  Soak and Diurnal
Area
Class
B
B
C
C
Vapor
Hot
source
Soak
Diurnal
Hot
Soak
Diurnal
Curve Fit
0.01557
0.00736
0
0
* RVP
* RVP
.014421 *
.006818 *
2 - 0
2 - 0
RVP2
0.
RVP2
0.
.1671 *
.0790 *
RVP
RVP
- 0.16248 *
5520
- 0.07682 *
2610
+ 0.
+ 0.
RVP
RVP
5399
2553
+
+
2 .
Phase II Nonexhaust VOC Emissions Model
     The Phase II Complex Model takes effect in the year 2000.
The equations giving nonexhaust VOC emissions for the Phase II
                                154

-------
Complex Model were derived in an analogous manner to that for the
Phase I model, with three main differences: MOBILE5a was used
instead of the July 11, 1991 version of MOBILE4.1, an enhanced I&M
program was assumed instead of a basic I&M program,  and emissions
from light-duty trucks were considered in addition to emissions
from passenger cars.  Thus nonexhaust VOC emissions were modeled
through curve fits to data produced by the MOBILE5a emissions
model (the input file to MOBILE5a is located in Appendix IV).
Again,  the year 2015 was chosen as the target year to force
MOBILE5a to consider only 1990 engine technology and evaporative
emission control systems.  And as discussed in Section III, an
enhanced I&M scenario was assumed to apply to Phase II.

     The Phase II Complex Model for nonexhaust emissions was based
on the MOBILE5a results for both light-duty cars and trucks
because sufficient data on light-duty truck exhaust emission
effects of fuel modifications now exist to allow the exhaust
emission portion of the Complex Model to cover both vehicle types.
When the simple model was developed, very few light-duty trucks
had been tested.  Emissions from heavy-duty gasoline trucks
(HDGTs)  have been ignored for two reasons: HDGTs comprise only 3
percent of the vehicle miles travelled by the in-use fleet  (as
estimated by MOBILE5a), and EPA does not have sufficient data to
describe the exhaust emissions performance of HDGTs as a function
of fuel parameters.  To maintain a consistent basis for both the
exhaust and nonexhaust models, EPA chose to exclude heavy-duty
gasoline trucks from the baseline emissions inventories and
emission models for both exhaust and nonexhaust emissions.

     To assess the relationship between the Reid vapor pressure of
a given fuel and nonexhaust VOC emissions, MOBILE5a was run for
the year 2015 under an enhanced I&M scenario with fuel RVPs
ranging from 6.5 to 9.9.  Separate results for Class C and Class B
areas were produced.  Each set of results gave emission factors in
terms of total grams of VOCs per mile, categorized by vehicle type
(LDV, HDV, etc.) and emission type  (refueling, evaporative, and
running losses).  Table IV-24 shows the MOBILESa results that were
used to develop the Phase II Complex Model for nonexhaust VOC
emissions.
                                155

-------
                            Table  IV-24
                Nonexhaust  Emissions  from MOBILE5a


Class

C

















B



















RVP

9.9
9.7
9.5
9.3
9.1
8.9
8.7
8.5
8.3
8.1
7.9
7.7
7.5
7.3
7.1
6.9
6.7
6.5
9.9
9.7
9.5
9.3
9.1
8.9
8.7
8.5
8.3
8.1
7.9
7.7
7.5
7.3
7.1
6.9
6.7
6.5
VOC emissions in grams per mile:
Refueling
Losses
LDGV
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.04
0.04
0.04
0.04
0.04
0.04
0.04
0.04
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.04
0.04
0.04
0.04
0.04
0.04
0.04
0.04
LDGT
0.07
0.07
0.07
0.07
0.07
0.07
0.06
0.06
0.06
0.06
0.06
0.06
0.06
0.06
0.06
0.05
0.05
0.05
0.07
0.07
0.07
0.07
0.07
0.07
0.06
0.06
0.06
0.06
0.06
0.06
0.06
0.06
0.06
0.05
0.05
0.05
Running
Losses
LDGV
0.58
0.51
0.44
0.41
0.39
0.37
0.34
0.32
0.30
0.28
0.26
0.24
0.21
0.19
0.17
0.15
0.14
0.14
0.66
0.58
0.49
0.47
0.44
0.42
0.39
0.36
0.34
0.31
0.29
0.26
0.24
0.21
0.18
0.16
0.15
0.15
LDGT
0.28
0.25
0.23
0.21
0.21
0.20
0.19
0.19
0.18
0.17
0.16
0.16
0.15
0.14
0.13
0.13
0.12
0.12
0.29
0.26
0.23
0.23
0.22
0.21
0.21
0.20
0.20
0.19
0.18
0.18
0.17
0.16
0.16
0.15
0.15
0.15
Evaporative
Losses
LDGV
0.23
0.21
0.19
0.18
0.17
0.16
0.14
0.14
0.13
0.12
0.11
0.10
0.09
0.09
0.08
0.08
0.08
0.07
0.29
0.26
0.24
0.22
0.20
0.19
0.18
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.10
0.09
0.08
0.08
LDGT
0.24
0.22
0.19
0.17
0.15
0.15
0.14
0.13
0.13
0.12
0.11
0.10
0.09
0.09
0.08
0.08
0.08
0.08
0.31
0.27
0.24
0.23
0.21
0.19
0.18
0.16
0.15
0.14
0.13
0.12
0.12
0.11
0.10
0.09
0.09
0.08
     The default values from MOBILESa for the percent of vehicle
miles traveled  (VMT)  by each vehicle type was used to produce a
weighted average of the results for light-duty cars and trucks.
In the MOBILESa output, these values are listed as 58.2% for
light-duty gasoline vehicles (LDGVs) and 29.3% for light-duty
                                156

-------
gasoline trucks (LDGTs),  with the remainder consisting of diesel
vehicles, heavy-duty gasoline vehicles, and motorcycles.
Normalizing these two values, the relative weighting of VMT
becomes 66.5 percent for light-duty gasoline cars and 33.5 percent
for light-duty gasoline trucks.  These weightings were applied to
the values in Table IV-24 above.  The resulting values are shown
in Table IV-25.
                                157

-------
                            Table  IV-25
                Nonexhaust  Emissions  from MOBILE5a

Class
C

















B


















RVP
9.9
9.7
9.5
9.3
9.1
8.9
8.7
8.5
8.3
8.1
7.9
7.7
7.5
7.3
7.1
6.9
6.7
6.5
9.9
9.7
9.5
9.3
9.1
8.9
8.7
8.5
8.3
8.1
7.9
7.7
7.5
7.3
7.1
6.9
6.7
6.5
VOC emissions in grams per mile:
Refueling
Losses
0.0567
0.0567
0.0567
0.0567
0.0567
0.0567
0.0533
0.0533
0.0533
0.0533
0.0467
0.0467
0.0467
0.0467
0.0467
0.0433
0.0433
0.0433
0.0567
0.0567
0.0567
0.0567
0.0567
0.0567
0.0533
0.0533
0.0533
0.0533
0.0467
0.0467
0.0467
0.0467
0.0467
0.0433
0.0433
0.0433
Running
Losses
0.4795
0.4229
0.3697
0.3430
0.3297
0.3131
0.2898
0.2765
0.2598
0.2432
0.2265
0.2132
0.1899
0.1733
0.1566
0.1433
0.1333
0.1333
0.5361
0.4728
0.4029
0.3863
0.3663
0.3497
0.3297
0.3064
0.2931
0.2698
0.2532
0.2332
0.2166
0.1933
0.1733
0.1567
0.1500
0.1500
Evaporative
Losses
0.2333
0.2133
0.1900
0.1767
0.1633
0.1567
0.1400
0.1367
0.1300
0.1200
0.1100
0.1000
0.0900
0.0900
0.0800
0.0800
0.0800
0.0733
0.2967
0.2633
0.2400
0.2233
0.2033
0.1900
0.1800
0.1600
0.1500
0.1400
0.1300
0.1200
0.1133
0.1033
0.1000
0.0900
0.0833
0.0800
     A numeric polynomial regression was carried out for each type
of emissions in Table IV-25 to produce a relationship giving VOC
emissions in grams per mile as a function of the Reid vapor
pressure in pounds per square inch.  This resulted in separate
equations for Class B and Class C areas, as shown in Table IV-26.
                                158

-------
                            Table  IV-26
 Nonexhaust Emissions Regressions as a Function of RVP - MOBILESa
Area
Class
B
B
B
C
C
C
Vapor source
Refueling
Running
Evaporative
Refueling
Running
Evaporative
Curve Fit
0.004767 * RVP + 0.011859
0.017768 * RVP2 - 0.18746 * RVP +
0.61457
0.014039 * RVP2 - 0.17075 * RVP +
0.60037
0.004767 * RVP + 0.011859
0.016169 * RVP2 - 0.17206 * RVP +
0.56724
0.010853 * RVP2 - 0.13346 * RVP +
0.48423
R2
0.908
0.983
0.996
0.908
0.985
0.993
     As stated above, evaporative emissions within the MOBILE
emissions model include both hot-soak and diurnal emissions.  The
ratio of hot soak to diurnal emissions is specific to the area
class.  For Class B areas, the average weighting of evaporative
emissions over all vehicle ages is 47.4 percent hot soak and 52.6
percent diurnal.  For Class C areas,  the average weighting of
evaporative emissions over all vehicle ages is 56.0 percent hot
soak and 44.0 percent diurnal.  These values differ from the
MOBILE4.1 values because of 1) the changes to the diurnal
calculations in the MOBILE model that were made when version 5a
was being developed, and 2) the change in I/M assumption in Phase
II with respect to Phase I.  Thus the equations for hot soak and
diurnal emissions can be derived from the equations for
evaporative losses shown in Table IV-26.  The results are shown in
Table IV-27.
                                159

-------
                            Table  IV-27
           Evaporative Emissions Breakdown for MOBILESa
Area
Class
B
B
C
C
Vapor source
Hot Soak
Diurnal
Hot Soak
Diurnal
Curve Fit
0.006654 * RVP2 - 0.08094
0.28458
0.007385 * RVP2 - 0.08981
0.31580
0.006078 * RVP2 - 0.07474
0.27117
0.004775 * RVP2 - 0.05872
0.21306
* RVP +
* RVP +
* RVP +
* RVP +
3.    Derivation of Nonexhaust Toxics Equations

     The MOBILE model was used to determine the relationship
between RVP and nonexhaust VOC emissions because MOBILE is based
on the most comprehensive collection of data on this subject.  The
MOBILE model would therefore be the most natural source of
information on the relationship between fuel parameters and
nonexhaust emissions of toxic compounds.  However,  the MOBILE
model only provides emission estimates for VOC, NOx, and carbon
monoxide.  Therefore, it was necessary for the EPA to develop the
nonexhaust toxics portion of the Complex Model in a manner
different from that used to develop the nonexhaust VOC model.

     General Motors  (GM) used their proprietary vapor equilibrium
model to predict the vapor composition above a pool of fuel
(conditions simulating a vehicle's fuel tank).  The GM model
provided separate estimates for refueling, diurnal, and running
losses using a fuel matrix consisting of 12 fuels.   These fuels
are shown in Table IV-28.  Fuels 1, 2, and 3 are the base fuels
that were used to produce the subsequent blends.
                                160

-------
                            Table  IV-28
                     GM Tank Vapor Model  Fuels
Fuel #
I
2
3
4
5
6
7
8
9
10
11
12
Oxygenate, vol%
none
none
none
MTBE, 11%
MTBE, 11%
MTBE, 11%
EtOH, 10%
EtOH, 10%
EtOH, 10%
EtOH, 10%
EtOH, 10%
EtOH, 10%
Type of blend
N/A
N/A
N/A
splash
splash
splash
splash
splash
splash
volatility-matched
volatility-matched
volatility-matched
RVP
7
8
8.7
7.46
8.36
9.03
8.20
9.13
9.82
7
8
8.7
     The fuel benzene content was kept constant at 1 percent by
volume in all twelve test gasolines.  Different benzene levels
were not tested since it is widely accepted that nonexhaust
benzene emissions will be proportional to fuel benzene content.
Match-blending was carried out by adjusting the concentrations of
n-butane and toluene.

     EPA believes that fuel tank vapor composition adequately
represents the composition of in-use emissions since nonexhaust
VOC emissions are dominated by emissions from vehicles which would
fail a pressure or purge test.  On such "failing" vehicles, the
canister has little effect on the composition of emissions,
resulting in emissions with compositions essentially identical to
their tank composition.  Emissions from passing vehicles might
have compositions which would differ from the composition of
failing vehicle fuel tank vapors.  However, nonexhaust emissions
from passing vehicles comprise only a small fraction of the total
nonexhaust emission inventory.  For this reason, EPA believes it
appropriate to base its model of the composition of nonexhaust
emissions on the composition of nonexhaust emissions from failing
vehicles.

     The mass percent of benzene in the refueling, diurnal, and
running loss vapors as predicted by their tank vapor model was
provided directly by General Motors.  These values are shown in
Table IV-29.  The range of temperatures used to obtain these
results was 90 °F for refueling emissions,  72-96 °F for diurnal
emissions, and 90-115 °F for running loss  emissions.   Since the
same temperature was used for both hot soak and running losses,
the results for these two categories of emissions were identical
and only the running loss results are shown below.
                                161

-------
                            Table  IV-29
                    GM Tank Vapor  Model  Results
Fuel #
I
2
3
4
5
6
7
8
9
10
11
12
Mass percent benzene in vapors:
Refueling
0.828
0.742
0.690
0.722
0.660
0.620
0.720
0.654
0.612
0.829
0.736
0.683
Diurnal
0.815
0.731
0.679
0.711
0.650
0.611
0.714
0.648
0.607
0.823
0.730
0.676
Running Loss
0.884
0.800
0.748
0.769
0.709
0.670
0.757
0.695
0.656
0.859
0.772
0.721
     The values in Table IV-29 indicate that the addition of MTBE
to gasoline lowers the mass percent of benzene in the vapor phase,
while the addition of ethanol has no effect on the mass percent of
benzene as compared to that for the base gasoline alone.  To
quantify the change in mass percent benzene for MTBE blends,
linear curve-fitting was applied to the base fuel and the MTBE
blend.  Table IV-30 shows the resulting equations giving the mass
percent benzene in the nonexhaust vapors for an unoxygenated
gasoline (fuels 1, 2, and 3 in Table 1-20), and an 11% MTBE blend
(fuels 4, 5, and 6 in Table 1-20) .

                            Table IV-30
              Linear Regressions of GM Model Results
Oxygenate
none
none
none
MTBE
MTBE
MTBE
Vapor source
Refueling
Diurnal
Running loss
Refueling
Diurnal
Running loss
Equation giving mass
percent benzene:
1.3972 - 0.081507 * RVP
1.3758 - 0.080274 * RVP
1.4448 - 0.080274 * RVP
1.2073 - 0.065186 * RVP
1.1868 - 0.063921 * RVP
1.2400 - 0.063258 * RVP
R2
0.998
0.999
0.999
0.998
0.998
0.999
     In an effort to simplify the set of equations in Table IV-30,
the equations giving the mass percent benzene for MTBE blends were
combined with the equations giving the mass percent benzene for
unoxygenated gasoline.  This was done by assuming that the slope
                                162

-------
of the equation for a clear gasoline was approximately equal to
the slope of the equation for an MTBE blend for a given vapor
source.  Thus the change in mass percent benzene due to the
addition of MTBE to a clear gasoline can be assumed constant over
the range of applicable RVPs.   The equations from Table IV-30 were
evaluated at 8.0 RVP (the approximate center of the range of RVPs
for fuels 1-6 in Table IV-28),  and the resulting mass percent
benzene values for MTBE blends were subtracted from the mass
percent benzene values for clear gasoline to obtain a "constant"
difference.  These results are shown in Table IV-31.

                            Table  IV-31
        Differences in Regression Constants Between Blends
Vapor
Source
Refueling
Diurnal
Running
Hot soak
Mass % benzene in
vapors for clear
gasoline at 8.0 RVP
0.7451
0.7335
0.8028
0.8024
Mass % benzene in
vapors for 11 vol%
MTBE blend at 8 . 0 RVP
0.686
0.6756
0.7338
0.7340
Change in mass %
benzene with the
addition of MTBE to
a clear gasoline
- 0.0591
- 0.0579
- 0.0684
- 0.0684
     An allowance can also be made for fuel MTBE contents of other
than 2.0% oxygen by weight (11% MTBE by volume)  by multiplying the
last column of Table IV-31 by the ratio of MTBE content for the
target fuel and a fuel with 11% MTBE.  The resulting equations,
giving the mass percent of benzene in the nonexhaust emissions for
any MTBE-oxygenated or unoxygenated blend, are shown in Table IV-
32.  Note that in all equations below, "MTBE" is the weight
percent of oxygen in the form of MTBE in the fuel ("MTBE" would
necessarily be zero for any oxygenate other than MTBE).

                           Table IV-32
                   Revised GM Model  Regressions
Vapor source
Refueling
Diurnal
Running loss
Hot soak
Equation
1
1
1
1
.3972 -
.3758 -
.4448 -
.4448 -
[0.
[0.
[0.
[0.
giving
0591
0579
0684
0684
*
*
*
*
mass
MTBE/2
MTBE/2
MTBE/2
MTBE/2
percent
.0]
.0]
.0]
.0]
- 0
- 0
- 0
- 0
benzene
.081507
.080274
.080274
.080274
.
*
*
*
*

RVP
RVP
RVP
RVP
     Two additional steps must be taken to convert the equations
in Table IV-32 to equations that provide the nonexhaust emissions
of benzene in grams/mile.  First, changes in the benzene content
of a fuel are expected to result in a proportional change in the
                                163

-------
benzene fraction of nonexhaust benzene emissions.  To account for
fuel benzene contents other than the 1 vol% benzene that was used
to establish the results in Table IV-29,  the equations giving the
mass percent benzene (Table IV-32)  should be multiplied by the
volume percent benzene in the fuel.   Second, to obtain benzene
emissions in grams/mile for each vapor source,  the equations
giving the mass percent benzene must be 1)  converted to mass
fraction benzene by dividing by 100, and 2) multiplied by the
total VOC emissions in grams/mile for that vapor source.  The
final equations are shown in Table IV-33.

                            Table  IV-33
        Phase  II  Complex Model  Nonexhaust Benzene Emissions
Vapor source
Refueling
Diurnal
Running loss
Hot soak
Equation
(BZV%
(BZV%/100)
(BZV%
(BZV%
giving benzene
/100) *
MTBE/2
* DVOC
/100) *
MTBE/2
/100) *
MTBE/2
RFVOC
.0] -
*
0
emissions in grams/mile:
(1.3972
.081507 *
* (1.3758 - [0.
0.080274 * RVP)
RLVOC
.0] -
HSVOC
.0] -
*
0
*
0
(1.4448
.080274 *
(1.4448
.080274 *
- to
RVP)
0579
- to
RVP)
- to
RVP)
.0591
*

* MTBE/2. 0]
.0684
.0684
*
*


BZV%  =   Volume percent of benzene in the fuel
RFVOC =   Total grams per mile of VOC from refueling emissions
DVOC  =   Total grams per mile of VOC from diurnal emissions
RLVOC =   Total grams per mile of VOC from running losses
HSVOC =   Total grams per mile of VOC from hot soak emissions

The values for RFVOC, DVOC, RLVOC, and HSVOC for the Phase I and
II Complex Models were presented in Sections B.I and B.2.

4.   Validation of Nonexhaust Toxic Emissions Model

     Southwest Research Institute was contracted by EPA in July,
1992 to determine how fuel volatility, as measured by the vapor
pressure, is affected by the addition of various oxygenates in
gasoline blends.  As part of this work,  SwRI completed a gas
chromatograph speciation analysis of a subset of the fuels that
were tested and their associated vapors.  From this speciation,
the mass percent benzene in the vapors and the volume percent
benzene in the fuel can be computed and compared to the nonexhaust
toxic emissions model described above.  Thus, despite the fact
that the GM tank vapor model is proprietary, the results of the
study carried out by SwRI independently validate and substantiate
the nonexhaust benzene correlations described above.
                                164

-------
     Table IV-34 shows the properties for five fuels whose vapor
pressure was measured at both 100 and 130°F.  The aromatics
content of fuels 3, 4, and 5 was increased through the addition of
toluene or 1,3,5-trimethylbenzene as indicated under the aromatics
content.  Note that the presence of ethanol in fuels 2, 3, and 5
is assumed to have no effect on the mass percent benzene in the
vapors  (see description of Table IV-29 above).

                            Table IV-34
                        SwRI Contract Fuels
Fuel
#
1
2
3

4

5
Aromatics
content
20%
20%
35% (toluene)
35%
( t rimethylben
zene)
35%
( t rimethylben
zene)
RVP
8.17
9.10
8.01

6.93

8.05
Oxygenate,
vol%
none
EtOH, 10%
EtOH, 10%

none

EtOH, 10%
BZV% at
100 Deg F
1.75
1.74
1.36

1.40

1.36
BZV% at
130 Deg F
1.70
1.60
1.37

1.40

1.36
     Using the volume percent benzene in the fuel and the RVP
levels for each of the five fuels shown in Table IV-28,  the
nonexhaust toxic emissions model described above can be used to
predict the mass percent benzene in refueling,  diurnal,  running
loss, and hot soak emissions.  These estimates can then be
compared to the mass percent benzene measured in the vapors by
SwRI.  These measured values do not represent specific
measurements for either refueling,  diurnal, hot soak, or running
losses.  However, since the difference in emission levels for
these four sources results only from the differences in
temperatures under which the GM vapor model was run, the values
measured by SwRI can be appropriately compared to all four
sources.  Tables IV-35 and IV-36 show these values for fuel
temperatures of 100°F and  130°F,  respectively.
                                165

-------
                            Table  IV-35
           Comparison of Model to Measurements at 100°F

Fuel
I
2
3
4
5
Emissions Model benzene mass % prediction:
Refueling
1.28
1.14
1.02
1.17
1.01
Diurnal
1.26
1.12
1.00
1.15
0.99
Running Loss
1.38
1.24
1.09
1.24
1.08
Hot Soak
1.38
1.24
1.09
1.24
1.08
SwRI
benzene
1.28
1.23
1.00
1.34
1.03
                            Table  IV-36
           Comparison of Model to Measurements at 130°F

Fuel
1
2
3
4
5
Emissions Model benzene mass % prediction:
Refueling
1.24
1.05
1.02
1.17
1.01
Diurnal
1.23
1.04
1.01
1.15
0.99
Running Loss
1.34
1.15
1.10
1.25
1.08
Hot Soak
1.34
1.15
1.10
1.25
1.08
SwRI
benzene
1.39
1.11
1.16
1.33
0.92
     Overall, the measured values for mass percent benzene given
by Southwest Research Institute differ from the values predicted
from the toxic emissions model by an average of +2.5 percent and a
maximum of 16.7 percent.  The average value implies that the
nonexhaust model for benzene tends to underestimate the true
emissions.  However, the repeatability of the RVP measurements in
the SwRI study was ±0.16 psi, equivalent to a percent change in
benzene emissions of ±3.8 percent (as calculated from the Phase
II, Class C nonexhaust equations).   Thus the bias implied by the
average difference of +2.5 percent appears to be within the
measurement accuracy of the study.   The data in Tables IV-35 and
IV-36 also imply that, for any given fuel, the model predictions
for mass percent benzene will differ from measured values by an
average of no more than -5.2 percent (95 percent confidence
limit).   EPA has interpreted these results to suggest that the
correlations derived from the GM tank vapor model are sufficiently
accurate to form the basis for EPA's model for nonexhaust benzene
emissions.

5.   Response to Comments

     The Agency received a number of comments on the accuracy of
the nonexhaust benzene model in light of an analysis of more
                                166

-------
recent data.  This analysis suggested that the fraction of benzene
in nonexhaust vapors should be higher than that predicted by the
GM model results.  Specifically,  the analysis of hot soak and
diurnal heat build data showed that the weight fraction of benzene
in the vapor is a function of the total emissions.  At extremely
low emission rates the fraction of benzene in the vapor is quite
high.  As the emission rate increased,  however, the benzene weight
fraction was found to asymptotically approach a lower limit.  The
lower limit for hot soak emissions was significantly higher than
predicted by the GM model.  The lower limit for the diurnal data
showed a close correlation to the GM model.

     In response to these comments, the Agency has identified two
areas in which the requirements of the reformulated gasoline
program were compromised in the nonexhaust benzene study offered
to it.  These two areas are described below:

•    The Act requires that the impact of a fuel change on
     emissions must be restricted to 1990 technology type vehicles
     for the purposes of the reformulated gasoline and anti-
     dumping programs.   The Agency has determined that carburetted
     vehicles represent only a small proportion (<2%)  of the 1990
     fleet.  However, a large fraction of the submitted heat build
     data included carburetted vehicles.   Thus the Agency has
     determined that the conclusions reached as part of the
     submitted study are not compatible with either the
     reformulated gasoline or anti-dumping programs.

     The high proportion of carburetted vehicles in the submitted
     test data is evident from the fact that carburetted vehicles
     typically have higher hot soak emission rates than fuel-
     injected vehicles.  The primary source of hot soak emissions
     from carburetted vehicles is the carburetor bowl which is
     extremely hot and may completely evaporate the fuel.   The
     primary source of hot soak emissions from fuel-injected
     vehicles is vapors created in the fuel tank at a much lower
     temperature.  Therefore the high emission rate data is
     probably dominated by carburetted vehicles which may consist
     of whole fuel vapors.  The submitted data and analysis for
     hot soak emissions support this conclusion since the weight
     fraction of benzene in the vapor approaches the fuel weight
     fraction as the emission rate increases.

•    The EPA has determined that the low emission rate data for
     fuel-injected vehicles is not representative of in-use
     emissions.  A large fraction of these emission may be caused
     by background emissions now characterized as resting loss
     emissions.  Resting losses were not included in the VOC
     baseline and so should not be included in the toxic baseline.
     The source of these resting losses has not been determined,
     but are thought to originate from permeation of fuel lines


                                167

-------
     and tanks.  Permeation emissions may be characterized by a
     high benzene weight fraction.

Based on its review of the submitted data and analyses on the
benzene fraction of nonexhaust vapors, the Agency has determined
that the General Motors vapor equilibrium model remains the most
appropriate and accurate source for developing the nonexhaust
benzene portion of the Complex Model.

     A number of commenters took issue with the Agency's proposed
effect of individual oxygenates on nonexhaust benzene emissions.
However, the conclusions reached by the different commenters were
contradictory.  One commenter pointed to Auto/Oil data which
indicated that MTBE increases rather than decreases nonexhaust
emissions of benzene.  Another commenter supported the suppression
of benzene emissions by MTBE, but indicated that recent
(unavailable)  data supported a similar effect for the oxygenates
ETBE and TAME.  Given the assailable nature and meager amount of
data on the effects of oxygenates on nonexhaust benzene emissions,
the Agency has determined that the Simple Model approach to the
development of a nonexhaust benzene model is the most appropriate
approach for the Complex Model.


C.    Determining Emission Performances

1.    Definition of Performance

     Section 211(k) of the Clean Air Act requires that refiners
make gasoline that results in a decrease in emissions of 15
percent (in Phase I of the reformulated gasoline program)  over the
emissions produced from baseline gasoline.  Thus the CAA
identifies the measure of "cleanliness" of a given fuel by the
percentage difference between its emissions and emissions from the
baseline gasoline. For the purposes of this rulemaking, therefore,
the performance of a given fuel is defined as the percent change
in the vehicle emissions that would occur if baseline gasoline
were to be replaced with the given fuel in the fuel tank of a
typical 1990 vehicle.  The definition of the baseline gasoline
depends on the season as described in Section III.A.

     The Complex Model is designed to estimate the effects of
various fuel parameters on emissions of VOC, NOx, and toxic
compounds.  If the Complex Model performance of a pollutant is
given by Y and the fleet average emissions for that same pollutant
in grams per mile terms is given by X, then performance of a
target fuel t as compared to a baseline fuel b is calculated from

     (C.I)     {[X(t) - X(b)]/X(b)} * 100 = Y(t)

The Complex Model can provide separate emissions performance
estimates for three main categories of pollutants: total VOC, NOx,

                                168

-------
and total toxics.  Each category includes emissions from one or
more sub-categories.  Thus the X term in Equation (C.I)  is the sum
of the contributions from various sources:

                            Table IV-37
          Contributing  Sources to Each Pollutant  Category
Pollutant
Category
VOC
NOx
Toxics
Contributing Sources
Exhaust VOC
Non- exhaust VOC
Exhaust
NOx
Exhaust benzene
Non- exhaust benzene
Exhaust 1, 3 -butadiene
Exhaust acetaldehyde
Exhaust formaldehyde
Exhaust POM
Note that the non-exhaust portion of the Complex Model contains
equations for only benzene and VOC,  since other pollutants are
found only in exhaust emissions.
2 .
Exhaust Emissions Calculations
     The Complex Model can provide both performance (i.e.,
percentage change)  and mass (i.e., gram per mile)  estimates of the
emissions from a target fuel as compared to a baseline fuel.  The
calculations for each pollutant in the exhaust model are performed
initially in terms of percentage change from baseline levels,
while the non-exhaust model's calculations are performed initially
in terms of grams per mile.  To obtain total performance estimates
as per Equation  (C.I), both exhaust and non-exhaust estimates must
be expressed in terms of grams per mile.  This section focuses
exclusively on the derivation of percentage change estimates for
exhaust emissions and their conversion to gram per mile estimates.
Section C.3. provides details on how this method for determining
emissions performance, though different in form from that proposed
in the February 1993 NPRM  [58 FR 17175] , is mathematically
identical to the originally proposed method.  Note that the
derivations shown in this and the following sections are based
upon the raw unconsolidated version of the final Complex Model,
but the conclusions are equally applicable to the model's
consolidated form.

     The process used to develop the exhaust emission Complex
Model resulted in separate equations for each technology group and
emitter class for all six exhaust pollutants.  Higher emitting
                                169

-------
vehicles, however, were not categorized by technology group for
the reasons discussed in Section A. 2.  Hence there are nine
distinct equations for each pollutant, one for each of the eight
normal emitter technology groups and a ninth for higher emitters.
Each of these nine equations relates fuel parameters to emissions
and has the following form:

      (C.2)      E± = C± * exp(fi(fuel parameters))

where E is emissions in grams per mile, C is a proportionality
constant of unknown magnitude, and the subscript i refers to a
specific vehicle group.   As described in Section A, the function
fA  is  a polynomial which can contain linear,  squared,  and second-
order interactive terms.

     The value of the proportionality constant C± is unknown
because the statistical development process used to generate the
Complex Model did not include the use of intercept terms  (see
Section IV.A.4 for details).  Thus Equation  (C.2) can only be used
to estimate the relative magnitude of emissions.  For example,
Equation (C.2) can be evaluated for a specific target fuel and for
a baseline fuel, resulting in the following system of equations:

      (C.3)      E^b)      = C± *  exp (fi (baseline fuel parameters))
                         = C± *  exp(fi(b))

      (C.4)      E^t)      = Ci *  exp (fi (target fuel parameters))
                         = C± *  exp(fi(t))

where b refers to the set of baseline fuel parameter values and t
refers to the set of target fuel parameter values.  E(b)i and E(t)i
can be combined to obtain the percentage change in emissions that
would occur in a vehicle of vehicle group i if baseline gasoline
were replaced with the target gasoline.  Mathematically this
percentage change is computed as follows:

      (C.5)      Percentage change in emissions from vehicles of
               vehicle group i

               = 100 *  [E^t)  -  Ei(b)]/Ei(b)

               = 100 *  [Ciexp(fi(t))  - Ciexp(fi(b))]/[Ciexp(fi(b))]

               =  [exp(fi(t))/exp(fi(b)) - 1]*100

The final form of Equation  (C.5) no longer contains the
proportionality constant C±.   As a result,  the  percentage change
in emissions can be calculated directly from Equation  (C.5).

     To determine the fleet-average percentage change in
emissions,  the equations for the individual vehicle groups must be
combined.  Since each vehicle group contributes a different

                                170

-------
fraction of the total fleet-wide emissions, the individual
percentage change estimates for each vehicle group must be
weighted to reflect the group's contribution to the overall
emissions inventory.  The derivation of these weighting factors
was discussed in Section III.B.4 above.  Thus the total
performance estimate for a given pollutant is given by

     (C.6)     Y = £ {Wi *  [exp(fi(t))/exp(fi(b))  - 1]  * 100}

where the w±  are  the weighting  factors  given in Table  III-8 of
Section III.B.4.

     Once the target fuel's fleet-average performance estimate has
been determined,  it must be converted to a gram per mile basis
before being combined with the non-exhaust emission estimates.
This conversion involves multiplying the fleet-average estimate of
the percentage change in emissions by the baseline emission
estimates X(b) derived in Section III.B.  The conversion has the
following form, with X(t) representing the fleet-average emissions
from the target fuel in gram per mile terms and Y(t)  representing
the fleet-average performance of the target fuel  in terms of
percentage change from baseline:

     (C.7)     X(t) = X(b)  + [X(b) * Y(t)/100]

Fleet-average exhaust and non-exhaust emissions,  both expressed in
gram per mile terms, can be summed for each pollutant.  The total
percentage change in emissions can then be calculated using
Equation  (C.I).

     The calculations described in this section do not take into
account any linear extrapolations.  When linear extrapolation is
required, the calculation procedures described in this section
must be supplemented with other calculations.  Both the conditions
under which extrapolation is required and the associated
calculation procedures are covered in Section IV.D.


3.    Comparison of the Current Performance Calculation Method and
     the Method Proposed in the February 1993 NPRM

     The calculation of fleet-average performances for each
exhaust pollutant described above circumvents the need for vehicle
group-specific gram/mile estimates.  This change  was made to the
proposed method for calculating performances (as  laid out in the
DRIA for the February 1993 NPRM, 58 FR 17175) to  simplify both the
calculations themselves and the conceptual approach to the Complex
Model.   It was also made in response to comments  that expressed
confusion over the derivation of the proportionality constants
(the kj  used in  the February 1993 proposed model.  The method
described above is, in fact, mathematically equivalent to the
originally proposed method, except that the proportionality

                                171

-------
constants and the calculation of vehicle group-specific gram per
mile emission estimates have been eliminated.  This section
describes the mathematical conversion of the proposed method to
the final method described above.

     As described previously, the exhaust equations for all
pollutants have the form of an exponential of a polynomial
function of the eight fuel parameters.  Each such equation is
unique for every vehicle group and pollutant in the raw
(unconsolidated) form of the Complex Model.  In the February 1993
Notice of Proposed Rulemaking [58 FR 17175] ,  a procedure for
calculating gram/mile values for every vehicle group was outlined.
This procedure provided total (fleet average) gram/mile values
from the gram/mile values for individual vehicle groups through
the following equation:

     (C.8)     X = E kiWiexp(fi)

X    =    Fleet average emissions of any pollutant in g/mi
ki    =    Adjustment factor for vehicle group i
f.±    =    Polynomial function for vehicle group i, evaluated with
          the fuel parameter values for the target fuel
Wi    =    Weighting factor for vehicle group i (note that these
          weighting factors are not the same as the weighting
          factors wi described in Section III.B.4  but  instead are
          the normalized sales weighting for vehicle group i,
          unadjusted for differences in baseline emissions across
          vehicle groups.  Thus Wi retains  the  weighting factor
          definition given in the February NPRM.)

In Equation (C.8), kiWiexp(fi)  is the gram/mile contribution to the
total mobile source inventory from vehicle group i.  Equation
(C.8) can be written more simply as

     (C.9)     X(t) = E X±(t)

to show that the total, fleet average gram/mile estimate is made
up of the gram/mile estimates from the contributing vehicle
groups.  Replacing X with X(t) shows that X is a function of the
target fuel.

     As described in Section C.I above, the total percent change
in the emissions of any exhaust pollutant is given by

     (C.I)      {[X(t) - X(b)]/X(b)} * 100 = Y(t)

Substituting Equation  (C.9) into Equation  (C.I) yields

     (C.10)     [(E x±(t) - E X^bM/E X±(b)]  *  100 =  Y±(t)

where the subscript i denotes individual vehicle groups.  This
equation can be applied to the raw, refined,  or consolidated

                                172

-------
Complex Models; using  the  raw form of the Complex Model, for
example, the summations  in the above equation can be written as
follows :

               E X±(t)  =  X±(t)  + X±(t) + - + X10(t)

               E x±(b)  =  x,(b)  + X2(b) + - + X10(b)

where vehicle groups #1  through #9 are the normal emitter
technology groups, while vehicle  group #10 is the higher emitting
vehicles.  Equation  (C.10)  then becomes Equation (C.ll):

            X10(t))-(X1(b) +•••+ X10(b)) 1*100
                 + - + X10(b))                 = Y(t)
The terms in the numerator  can be grouped according to vehicle
group to yield Equation  (C.12) :

                  + (X2(t)  -  X2(b)) + •• 1*100
                 +  X10(b))                           = Y(t)
Equation  (C.12) can be  separated into ten separate equations, each
representing a given vehicle  group's contribution to the total
percent change in pollutant Y:
(C.13a)   [ (X.U) -X-^b) )/E X±(b)]  * 100 =  Contribution to total
                                         percent change in
                                         pollutant Y from vehicle
                                         group #1

(C.13b)   [ (X2(t) -X2(b) )/E X±(b)]  * 100 =  Contribution to total
                                         percent change in
                                         pollutant Y from vehicle
                                         group #2

and so on  for  (C.13c),  (C.13d),  etc.

At this point, combining Equations  (C.8)  and (C.9) gives

      (C.14)    X±(t) = kiWiexp(fi(t))

for the target fuel,  and

      (C.15)    X±(b) = kiWiexp(fi(b))

for the baseline  fuel.  The  adjustment factors ki are  defined as
follows :

^   =     Ci/exp(fi(b))
CA   =     Average emissions  of  pollutant Y from vehicle group i ;
           see EPA memorandum from Christian Lindhjem to Richard A.
           Rykowski,  "Effects of Oxygenates on Emissions," Jan.  7,

                                173

-------
          1992 and the DRIA  for  the  February 1993 NPRM for further
          details .

Substituting for ki in Equations  (C.14) and  (C.15)  results  in

      (C.16)    X±(t)       =     kiWiexp(fi(t))

                               CiWiexp(fi(t))/exp(fi(b))

      (C.17)    X±(b)       =     kiWiexp(fi(b))

                               CiWiexp(fi(b))/exp(fi(b))
Equation  (C.17) implies that  the  baseline values X^b)  are
constant.  Equation  (C.16)  implies  that  the target fuel values are
equal to the corresponding  baseline values multiplied by the ratio
of exp(fi(t))  to exp(fi(b)).  Using  vehicle group #1 as an example,
combine Equations  (C.13a),  (C.16),  and (C.17)  to obtain

      (C.18)    Contribution to  total percent change in pollutant Y
               from vehicle group #1
                    = 100*(X1(t) - x^bM/E X±(b)

                    = 100*[C1W1exp(f1(t))/exp(f1(b)) - C.WJ /E C^

                    = 100*(C1W1)*[exp(f1(t))/exp(f1(b)) - 1] A C^

                    = 100*exp(f1(t))/exp(f1(b))  - 1] * (C^) /E C^

Equation  (C.18) provides vehicle group #1 ' s  contribution to the
total percent change in pollutant  Y.   The  term
100* [exp (f-L (t) ) /exp (f-L (b) ) -1]  provides the  percent change  in
pollutant Y for vehicle group #1,  equivalent to Equation  (C.5) .
Thus if the form of Equation  (C.18) were duplicated for the
remaining vehicle groups and  all such equations were summed,  the
result would be Equation  (C.7).  Since Equations (C.18)  and (C.7)
are mathematically equivalent,  the method  proposed in the February
NPRM for calculating performances  is  mathematically equivalent to
the method that has been employed  for the  final Complex Model .

     Since the C± and Wi  are  constants, the term C1W1  is also a
constant, and  (C^J/E CiWi  is  the normalized  form of  C1W1 .  Thus the
final weighting factor wi for vehicle  group  #1  (not to be  confused
with the interim weighting factor  WJ   is defined  as
The final weighting factors WA were derived in Section  III.B.4
above .

                                174

-------
     Although the derivation of the vehicle group-specific percent
change equations outlined above uses the raw form of the final
Complex Model, the results are applicable to the refined,
consolidated, and uncentered forms of the Complex Model as well.
As discussed above, the technology group weightings for normal
emitters (Table IV-2)  were applied to the normal emitter vehicle
groups in the random balance approach to vehicle group
consolidation.  Doing so resulted in a single equation for normal
emitters, and the emitter class emission weightings from Table 4
were then applied directly to the two remaining equations for each
pollutant.   Thus the consolidated and uncentered models contain
two vehicle groups for each pollutant instead of the original
nine, one for normal emitters and one for higher emitters.  The
derivations described in this subsection apply regardless of the
number of vehicle groups in a given model.

4.   Response to Comments

     A number of commenters noted the disparity between the
quantity of data on higher emitting vehicles in the Complex Model
database and the large weighting that higher emitters are given in
the Complex Model, particularly for VOC and toxics.  Two options
were suggested for alternative treatments of higher emitters:

     1)   Assume that the emissions impact of fuel modifications
          on higher emitters is generally similar to the emissions
          impact of fuel modifications on normal emitters.  This
          assumption would imply that the statistical development
          of the Complex Model should proceed with a "totally
          unified approach," in which higher emitters would be
          modeled in the same manner as normal emitter technology
          groups.

     2)   Decrease the weighting assigned to higher emitter
          effects in the final Complex Model.

     The Agency's analysis of available data indicated that higher
emitters have fundamentally different responses to fuel
modifications than do normal emitters.  The EPA therefore
continued to treat higher emitters as a distinct population of
vehicles requiring different analysis techniques than those used
to account for normal emitter technology group distinctions.  For
this reason, EPA decided not to use the totally unified modeling
approach.

     The Agency acknowledges the disparity between the proportion
of higher emitter data and the weighting given the higher emitter
equations in the Complex Model.  However, the Agency does not
believe that the limited available data on higher emitters
justifies under-weighting emissions from such vehicles for the
following reasons:


                                175

-------
     •    The Agency has no basis for arbitrarily lowering the
          weighting for higher emitters to a level below the
          contribution of such vehicles to in-use emissions.
          Doing so would result in models which are inconsistent
          with EPA's knowledge of in-use emissions and with EPA's
          MOBILE models.

     •    All interested parties were informed of the scarcity of
          data on higher emitters and the large weight that higher
          emitters were to receive in the final Complex Model
          during the early stages of model development,  beginning
          with EPA's first public workshop on the Complex Model
          held January 21-22,  1992.  Despite the Agency's
          solicitation for additional higher emitter data, the
          only additional higher emitter data included in the
          Complex Model database were obtained through EPA-
          sponsored test programs.  The Auto/Oil program did test
          higher emitters but failed to stabilize their emissions
          prior to testing.  As discussed in Section A.I, EPA
          considers inclusion of test results from unstable
          vehicles to be inappropriate.

     •    The fleet-average ratio of normal emitter emissions to
          higher emitter emissions is based on an analysis of data
          used to develop the MOBILE emissions model.  Since this
          data is regarded as the most complete set of available
          data on the in-use mobile source emissions inventory,
          EPA considers any other basis for these weightings to be
          less reliable.

The Agency considers the weighting factors described and derived
above to provide the most appropriate allocation of the emissions
inventory to each emitter class and technology group.
D.    Range and Extrapolation of the Model

     Section C presented the procedure for calculating emission
performances with the Complex Model equations.  The current
section describes some of the limitations of those regression
equations by addressing two specific issues:

     •    The fuel parameter ranges within which the Complex Model
          equations can be applied.

     •    The methodology for extrapolating the Complex Model
          equations to cover fuels whose properties are not
          represented in the Complex Model database.

These limitations on the use of the regression equations arise
from the limitations of the Complex Model database.  In regions of

                                176

-------
the fuels space43 where  emissions data is limited or absent, the
regression equations may not  reliably reflect the response of in-
use vehicles to fuel modifications.   Limits  on fuel parameter
ranges and extrapolation techniques  are  two  methods that EPA has
chosen to compensate for the  limitations of  the Complex Model
database.

     This section is divided  into five parts.  Section D.I derives
and presents the limits of  the  data  core,  which is the subset of
the fuels space that contains the bulk of the data in the Complex
Model database.  The data core  delineates the range of fuel
parameters within which EPA considers the regression equations to
be most trustworthy since sufficient data from the Complex Model
database is available.  Section D.2  presents EPA's approach to
designating the valid range,  or the  range of fuel parameters
within which fuels can be evaluated  with the Complex Model.  Since
the data core is smaller than the valid  range,  EPA has chosen to
employ a type of extrapolation  to estimate emission performances
between these two ranges.   Sections  D.3,  D.4, and D.5 present the
extrapolation methodology in  three steps:

     1)   Section D.3 lays  out  the Agency's  theoretical and
          mathematical approach to linear extrapolation.

     2)   Section D.4 identifies the location of inappropriate
          extrema  (minimums and maximums)  which can be
          accommodated in the Complex Model  through the
          application of linear extrapolation.

     3)   Section D.5 applies the approaches outlined in Sections
          D.3 and D.4 to the  Complex Model regression equations
          and presents the  final extrapolation algorithm and
          equations.

     Both the fuel parameter  ranges  and  extrapolation
methodologies are independent of the particular season, program
phase, and area class scenarios under which  the Complex Model can
operate.  The specific extrapolation approach to the Complex Model
applies to both the reformulated gasoline program and the anti-
dumping program, but the limits on the fuel  parameter ranges over
which the Complex Model is  valid differ  for  the reformulated
gasoline and anti-dumping programs.

1.    Derivation of Data Core
     43
        The "fuels space" refers to the multi-dimensional universe in which


each axis is represented by a single fuel parameter.  A given point in the fuels

space specifies a set of fuel properties which represents a single, unique fuel.


                                177

-------
     As introduced above,  the data core is the  subset of the
multi-dimensional  fuels space that contains  the bulk of the data
in the Complex  Model  database.  It can be pictured as a multi-
dimensional box which is defined by upper and lower bounds on each
fuel parameter.  The  data core serves three  main purposes:

     1)    It provides information on the limitations of the
           Complex  Model,  which was derived from a database which
           did not  cover all possible fuel formulations.

     2)    It delineates the range for each fuel parameter within
           which the regression equations are most reliable.

     3)    It provides the locations in the multi-dimensional fuels
           space at which extrapolation should begin.44

Specification of the  data core is the first  step in translating
the limitations of the Complex Model database into the limitations
of the Complex  Model.

     For the February 1993 Notice of Proposed Rulemaking  [58 FR
17175], the data core was determined from a  visual examination of
the data in the Complex Model database.45  This  visual examination
used frequency  distributions of the values of individual fuel
parameters.  The frequency distributions were simply graphical
displays in which  the horizontal axis gave the  level of a given
fuel parameter  and the vertical axis gave the fractional frequency
of observations of those fuel parameter levels  in the Complex
Model database.  The  Agency used the graphical  displays to specify
the low and high values of each fuel parameter  which enclosed the
bulk of the data (see the February 1993 Draft Regulatory Impact
Analysis for further  details).

     The Agency has since conducted a more rigorous analysis of
the data core based on four main factors:

     •     Distribution of the data.
     •     Covariation between fuel parameters.
     •     Behavior of the Complex Model.
     44
         Extrapolation generally begins at the limits of the data core.


However,  extrema also play a role in determining the locations at which

extrapolation should begin, as described in more detail in Section D.4.

     45
         In the February 1993 NPRM, the data core was referred to as the valid


range of  the Complex Model, and indeed was  coincident with the valid range as

defined and specified in Section D.2 below.


                                 178

-------
     •    The fuel parameter ranges which encompass
          various percentages of the data.

     The distribution of the data is an extremely important factor
because regression equations tend to be most reliable in the
regions of the fuels space that contain the most data.  Ideally,
the data in the database would be evenly distributed  (i.e.,
homogeneous)  over a given range for each fuel parameter.  To the
extent that the database is not homogeneous, the regression
equations are that much less reliable.

     Homogeneity is particularly important when examining the ends
of the data ranges.  Since the ends of the data ranges typically
have fewer observations than more central portions of the data,
the reliability of the regression equations decreases as one
approaches those ends.  This tendency implies that the limits of
the data core should not be coincident with the maximum and
minimum values observed in the database since it is near such
extreme values where the reliability of the regression equations
decreases rapidly.

     Covariation between fuel parameters is an issue in
determining the limits of the data core because it can obscure
holes in the Complex Model database.  Some portions of the multi-
dimensional fuels space may appear to contain sufficient data when
in fact they do not.  For example, the bulk of the data for E200
may lie between 35 and 70 percent for a given data set, while most
E300 values for the data set range from 70 to 95 percent.
However, since fuels with high E200 values typically contain high
E300 values and vice-versa, not all combinations of E200 and E300
levels would be well represented in the data set.  For example,
fuels with high E200 values (e.g., 65 percent) and low E300 values
(e.g., 75 percent) would not be well represented in the data set,
although the E200 and E300 values individually fall within their
respective ranges in the data set.  Significant covariation
problems can be identified with graphical displays by plotting the
levels of one fuel parameter as a function of a second fuel
parameter.  Covariation can also be identified analytically by
determining how well one can predict levels of a fuel paramter if
one knows the level of a second fuel parameter.  For example, an
indication of the degree of covariation between two parameters can
be determined by calculating the R-squared value for a regression
with one fuel parameter as the dependent variable and a second
fuel parameter as the independent fuel variable.  High covariation
would result in high R-square values.

     If a significant correlation exists between two fuel
parameters, then the data core limits for one fuel parameter
depend on the level of a second fuel parameter.  Covariation has
the effect of changing the shape of the data core from a multi-
dimensional box to some other, more complex shape.  EPA's
examination of covariation in the Complex Model database revealed

                                179

-------
substantial covariation between the following fuel parameter
pairs:

     Benzene and Sulfur       Aromatics and E300
     Benzene and Oxygen       Olefins and Sulfur
     Aromatics and E200       Olefins and Benzene
     E200 and Oxygen          E300 and Sulfur
     E200 and E300

However, in no case did the covariation result in the complete
absence of emissions data from any portion of the multi-
dimensional fuels space.  Thus it was not necessary to specify the
data core with the upper and lower bounds on each fuel parameter
being a function of the values of another fuel parameter.
Instead, the upper and lower bounds for each fuel parameter could
be determined independently.

     The behavior of the Complex Model is also an important factor
in determining the extend of the data core.  When discussing the
model's behavior, the exponential form of the model alters the
traditional implications of first- and second-order terms.  Since
the statistical regressions employed to develop the exhaust
emission Complex Model used the natural logarithm of emissions as
the dependent variable  (see Section A.2), emissions are
proportional to the exponential of the resulting polynomials,  not
to the polynomials themselves.  As a result, emissions are not
first-  or second-order functions of the fuel parameters included
in the model in a strict mathematical sense.  However, the
deviation between the behavior of an exponential of a polynomial
and the behavior of the polynomial itself is significant only for
large numerical values of the independent variables (the fuel
parameters, in the case of the complex model).  In the Complex
Model,  only sulfur measurements ever exhibit such high values, and
then only under the anti-dumping program (as discussed in Section
D.2).   Therefore, the behavior of the polynomial functions which
are exponentiated in the Complex Model's exhaust emission
equations provides a good approximation to the behavior of the
actual  exponential equations that comprise the exhaust emission
Complex Model.  For this reason, the Complex Model exhaust
equations themselves will be referred to as being first- or
second-order in the discussions that follow based on the form of
the exponentiated polynomial with no loss of descriptive accuracy.

     With the foregoing discussion in mind, the behavior of the
Complex Model can provide additional guidance in defining the
limits  of the data core.  As stated in the introduction to this
section, one purpose of the data core is to provide the locations
in the multi-dimensional fuels space at which extrapolation should
begin,  since the regression equations become less reliable than
desired outside the data core.  Models which exhibit questionable
or counter-intuitive behavior are less reliable than models which
do not  exhibit such behavior.  As a result, the range of fuel

                                180

-------
parameters within which the regression equations are considered
reliable should be made more restrictive for those parameters for
which the complex model exhibits questionable or counter-intuitive
behavior.  In other words,  the size of the data core should be
reduced in such cases so as to restrict poorly-behaved equations
to smaller regions of the fuels space than well-behaved equations.

     As described in Section D.3,  all extrapolations beyond the
limits of the data core were linear extrapolations.  Since a
linear extrapolation of a linear relationship results in the
original linear relationship,  linear extrapolation of the Complex
Model regression equations only alters the second-order
relationships included in the Complex Model.   Thus the behavior of
pollutant models should influence the extent of the data core only
for fuel parameters which are present in second-order terms.  The
data core should be far less influenced by the Complex Model's
behavior for fuel parameters present only in linear terms.  The
second-order terms which are present in the Complex Model are
shown in Table IV-38.

                            Table  IV-38
               Second-order terms  in the Complex Model
Pollutant
VOC
NOx
Benzene
Butadiene
Acetaldehyde
Formaldehyde
Second-order terms in the
regression equation
E2002, E3002, ARO*E300
OLE2 , SUL2 , ARO2
none
none
none
none
     As described in Section D.4,  second-order terms may create
extrema which may not appropriately represent the effects of fuel
modifications for all possible fuel parameter values.  In some
cases the existence and location of extrema are artifacts of the
distribution of the data in the Complex Model database.  In such
cases, limiting the size of the data core helps exclude
inappropriate extrema from the model.

     Finally, the range which encloses a fixed percentage of the
data for each fuel parameter also helps determine the extent of
the data core.  Not only does such an analysis provide a less
qualitative approach to specifying the limits which enclose the
bulk of the data, it also provides information on the homogeneity
of data at the ends of the data ranges.  Table IV-39 shows the
results of such an analysis.  Table IV-39 includes five range


                                181

-------
restriction limits.  The 98% limits,  for example,  are determined
by allowing 1% of the data to fall below the "low" end of the
range and 1% of the data to fall above the "high"  end of the
range.  Similarly,  the 95% limits are determined by allowing 2.5%
of the data to fall below the low end and 2.5% to fall above the
high end.  Each limit is pollutant-specific for each parameter
since not all pollutants were measured for each fuel-vehicle
combination.  For example, 96% percent of the acetaldehyde
observations were based on fuels with aromatics levels between
11.5 and 45.9 vol%, but 96% of the VOC observations were based on
fuels with aromatics levels between 17.9 and 45.9  vol%.
                                182

-------
         Table  IV-39
Limits of the Bulk  of  the Data

Parameter

Aromatics





Benzene





E200





E300





Olef ins






Pollutant

Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
98% limits

Low
11.3
11.3
11.3
11.3
9.3
11.5
0.39
0.39
0.39
0.39
0.39
0.38
30.96
26.68
26.68
30.96
27.42
30.02
71.65
71.19
71.19
71.65
71.65
71.65
0.6
1
0.6
0.6
0
1

High
46 .7
46.7
46.7
46 .7
47 .8
47.3
2.3
2.3
1.8
2.3
2.4
1.8
67 .41
67.41
67.41
67 .41
67.8
67.41
97.28
97.28
97.28
97.28
97.28
97.28
21.8
21.8
21.8
21.8
21.9
21.8
97% limits

Low
11.3
11.3
11.3
11.3
9.8
17.9
0.39
0.4
0.4
0.39
0.39
0.39
31.72
30.96
30.96
31.72
30.02
30.96
71.99
71.65
71.65
71.99
71.99
71.99
1
1.1
1
1
0.2
1

High
46.6
46.4
46.6
46.6
46 .7
46.7
1.84
1 .84
1.8
1.84
2.3
1.8
66.87
66.87
66.87
66.87
67 .41
67.41
97.28
97.28
97.28
97.28
97.28
95.56
20.3
20.3
20.3
20.3
21.8
21.8
96% limits

Low
11.5
11.5
11.5
11.3
11.3
17.9
0.4
0.4
0.4
0.4
0.39
0.39
31.72
30.96
30.96
31.72
30.96
31.72
72.02
71.99
71.99
72.02
72.02
71.99
1.1
1.1
1.1
1.1
0.6
1.1

High
45.9
45.9
45.9
45.9
46 .7
45.9
1.8
1.8
1.8
1.8
1.9
1.8
66.87
66.87
66.87
66.87
67 .41
66.87
95.56
95.56
95.56
95.56
95.56
95.31
19.3
19.3
19.3
19.3
20.3
20.3
95% limits

Low
11.5
11.5
11.5
11.5
11.5
18.1
0.4
0.4
0.4
0.4
0.4
0.39
31.88
31.34
30.96
31.88
31.34
31.88
72.68
72.02
72.02
72.68
72.02
72.02
1.1
1.1
1.1
1.1
0.9
1.1

High
45.7
45.7
45.7
45.7
46 .4
45.9
1.8
1.8
1.8
1.8
1.84
1.7
65.74
65.74
65.74
65.74
66.87
66.87
95.31
95.31
95.31
95.31
95.31
94.5
18.5
18.3
18.5
18.5
20.3
20.3
94% limits

Low
11.5
12.7
11.5
11.5
12.7
18.1
0.4
0.4
0.4
0.4
0.4
0.4
34.27
31.72
31.72
34.27
31.72
32.65
72.68
72.68
72.68
72.68
72.68
72.02
1.1
1.1
1.1
1.1
1
1.1

High
45.7
44.3
44.3
45.7
45.9
45.7
1.8
1.8
1 .7
1.8
1.8
1 .7
65.37
65.37
65.37
65.37
65.74
65.74
95
94.87
95
95
94.87
93.87
18.3
18.3
18.3
18.3
19.3
19.3
             183

-------
Oxygen





RVP





Sulfur





Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
Acetaldehyde
Benzene
Butadiene
Formaldehyde
NOx
VOC
0
0
0
0
0
0
6.2
6.2
6.2
6.2
6.4
6.4
31
31
31
31
7
10
3.81
3 .81
3 .81
3.81
3.42
3 .42
9.7
9.59
9.59
9.7
12
12
371
371
371
371
466
466
0
0
0
0
0
0
6.2
6.2
6.2
6.2
6.4
6.4
31
31
31
31
7
12
3.81
3 .81
3 .81
3.81
3.37
3 .37
9.59
9.59
9.59
9.59
12
12
371
371
371
371
466
466
0
0
0
0
0
0
6.2
6.2
6.2
6.2
6.6
6.5
31
31
31
31
10
21
3.48
3 .42
3 .42
3.48
3.37
3 .37
9.55
9.55
9.55
9.55
11.5
11.7
371
371
371
371
466
466
0
0
0
0
0
0
6.4
6.4
6.4
6.4
6.6
6.6
31
31
31
31
11
31
3.42
3 .42
3 .42
3.42
3.37
3 .37
9.4
9.32
9.32
9.4
10.2
10.2
371
371
371
371
452
466
0
0
0
0
0
0
6.4
6.4
6.4
6.4
6.86
6.86
31
31
31
31
12
31
3.42
3 .37
3 .37
3.42
3.34
3 .37
9.32
9.2
9.3
9.32
9.9
9.9
364
364
364
364
443
452
184

-------
     Table IV-39 is particularly useful when evaluating the
limits of the data core for fuel parameters which are included in
second-order terms as shown in Table IV-38.  The second-order
effects decrease the reliability of the complex model's emission
estimates for fuels with levels of such fuel parameters that are
outside the bulk of the data.  Thus the distribution of E200 data
in Table IV-39 for VOC observations should play a stronger role
in setting the limits of the data core than the distribution of
E200 data for NOx or benzene, since nonlinear E200 terms are
present only in the VOC model.  The same considerations apply to
the other fuel parameters:  the distribution of VOC observations
should play a strong role in determining data core limits for
E300 and aromatics, while the distribution of NOx data should
play a strong role in determining the data core limits for
olefins, sulfur, and aromatics.

     RVP, oxygen, and fuel benzene content never appear in
second-order terms in the exhaust Complex Model.  This fact
suggests that the effects of these parameters can be extended via
the regression curve approach to extrapolation  (see Section D.3)
with relatively high confidence to parameter levels for which
emissions data is scarce or absent.  Thus the location of the
data core limits for these three parameters has no bearing on
their extrapolation.  To convert the ranges from Table IV-39 to a
more manageable form for these three fuel parameters, the range
limits for individual pollutants can be combined into a single
set of ranges for all pollutants.  The number of observations for
each pollutant in the Complex Model database can be used to
develop a weighted average across all pollutants for RVP, oxygen,
and fuel benzene limits.  The weightings are shown in Table IV-
40; useable observations are those observations were not deleted
as outliers or influential points during the model-building
process.

                          Table  IV-40
           Number of Useable Observations by Pollutant
Pollutant
Acetaldehyde
Benzene
1, 3 -Butadiene
Formaldehyde
NOx
VOC
Total
Number of observations
in the database
1785
1872
1821
1788
5246
4342
16,854
Percent of total
observations
10.59
11.11
10.80
10.61
31.13
25.76
100.00
     This analysis allows conversion of the values in Table IV-39
into a form that is more conducive to determining an appropriate
data core, as shown in Table IV-41.
                               185

-------
              Table IV-41
Restricted Limits  of  the  Bulk of the Data
Parameter
Aromatics
Benzene
E200
E300
Olef ins
Oxygen
RVP
Sulfur
Pollutant
NOx
VOC
Overall
VOC
VOC
NOx
Overall
Overall
NOx
98% limits
Low
9.3
11.5
0.39
30.02
71.65
0
0
6.31
7
High
47 .8
47.3
2.15
67.41
97.28
21.9
3 .59
10.98
466
97% limits
Low
9.8
17.9
0.39
30.96
71.99
0.2
0
6.31
7
High
46 .7
46.7
1.97
67 .41
95.56
21.8
3 .56
10.96
466
96% limits
Low
11.3
17.9
0.39
31.72
71.99
0.6
0
6.40
10
High
46 .7
45.9
1.83
66.87
95.31
20.3
3 .40
10.71
466
95% limits
Low
11.5
18.1
0.4
31.88
72.02
0.9
0
6.51
11
High
46 .4
45.9
1.79
66.87
94.5
20.3
3 .39
9.84
452
94% limits
Low
12 .7
18.1
0.4
32.65
72.02
1
0
6.66
12
High
45.9
45.7
1.76
65.74
93.87
19.3
3 .37
9.63
443
                  186

-------
     Table IV-41 contains both pollutant-specific limits and
common limits.  The pollutant-specific limits are taken directly
out of Table IV-39, and represent the data limits for pollutants
whose behavior includes second-order effects.  The common limits
are determined by weighting all the pollutant-specific data
limits according to the values in Table IV-40.  The common limits
therefore represent the data limits across all pollutants.  The
final limits of the data core are discussed separately for each
parameter.
Aromatics
Benzene:
E200:
Since second-order aromatics terms occur in both
the NOx and VOC models,  EPA chose to be cautious
in setting the aromatics limits of the data core.
Table IV-39 shows that the VOC limits are always
more restrictive than the NOx limits.  When moving
from the 98% VOC limit to the 97% VOC limit,  the
low end of the aromatics range jumps from 11.5 to
17.9 vol%, indicating that very few observations
are present between those aromatics levels.  The
low end remains stable at approximately 18 vol% as
the data limits become more restrictive.  Thus the
low end of the data core has been set at 18 vol%
aromatics.  Between the 96% and 94% limits, the
high end remains stable at approximately 46 vol%
aromatics.  Between the 96% and 97% limits,
however, the high end jumps from 45.9 to 46.7
vol%, indicating that the data set begins to
become sparse beyond 46 vol%.  This conclusion is
supported by the 98% high end limit.  Thus the
high end of the data core has been set at 46 vol%.

Since benzene is present in the benzene model only
in a linear term, EPA was less cautious in setting
benzene limits.  The low end appears to be fixed
at approximately 0.4 vol% across all examined
limits, so EPA set the low end of the data core at
0.4 vol%.  Between the 94% and 96% limits, the
high end is stable at approximately 1.8 vol%.  The
high end jumps to 1.97 vol% at the 97% limit and
then to 2.15 vol% at the 98% limit, indicating
that data become scarce above the 96% limit but
are relatively common below it.  Therefore the
high end of the data core for benzene has been set
at 1.8 vol%.

E200 is present in the VOC model in a second-order
term, so EPA based its E200 data core limits on a
cautious interpretation of the distribution of VOC
data.  The low end appears to change little
between the 95% and 96% VOC limits, indicating
that a clump of data exists for E200 values near
31.8 vol%.  However, the low end increases to

                187

-------
E300:
Olefins
Oxygen:
RVP:
32.65 vol% at the 94% limit and decreases to 30.96
vol% at the 97% limits; these jumps indicate that
data is not homogeneously distributed at the low
end of the E200 range.  Thus the low end of the
data core has been set conservatively at 33 vol%.
A similar inhomogeneity exists on the high end:
more high end data exists between the 95% and 96%
limits than exists between the 94% and 95% limits.
Thus the high end of the data core has been set
conservatively at 66 vol%.

E300 is present in two separate second-order terms
for VOC, so EPA based its E300 data core limits on
a cautious interpretation of the distribution of
VOC data.  The low end is relatively constant at
72 vol%, which indicates that a significant amount
of data exists at the extreme low end of the
range.   Thus the low end of the data core was set
at 72 vol%.  However, the high-end E300 limits
vary significantly across the five data limits
examined.  EPA therefore set the high end of the
data core conservatively at a level of 94 vol%.

A second-order olefins term is present in the NOx
model,  so EPA based its olefins data core limits
on a cautious interpretation of the distribution
of NOx data.  The low end decreases rapidly
between the 95% and 98% limits, but is relatively
stable between the 94% and 95% limits.  Since the
low end data begins thinning out dramatically
below 0.9 vol%, the low end of the data core was
set at 1.0 vol%.  On the high end, a data gap
exists between the 96% and 97% limits, but
significant data appear available between the 95%
and 96% limits.  Thus the high end was set
conservatively at 19 vol%.

The low end of the data distribution is fixed at
0.0 wt% for all five data limits.  The high end is
constant at approximately 3.4 wt% between the 94%
and 96% limits.  Above the 96% limit, the high end
data becomes thinner.  Therefore, the low end of
the data core was set at 0.0 wt% and the high end
was set at 3.4 vol%.

RVP is not involved in any second-order terms in
the Complex Model, so EPA was less cautious in
setting RVP limits for the data core.  Although a
clump of data exists between the 97% and 98%
limits for the low end, the data is much thinner
between the 94% and 97% limits.  The low end of
the data core was therefore set at 7 psi.  The

                188

-------
               high end of the data distribution averages
               approximately 10 psi across all pollutants and all
               data limit levels,  so the high end of the data
               core was set at 10  psi.

Sulfur:        Sulfur is present in second-order terms in the NOx
               model, so EPA based the  sulfur limits of the data
               core on the distribution of NOx data.  At the low
               end of the range, the measurement tolerance for
               sulfur is much larger than the difference between
               the values for the  various range limits.  Thus the
               low end of the data core was set at the average
               level of 10 ppm sulfur.   EPA retained the NPRM
               high end limit of 450 ppm since there appears to
               be a significant amount  of data near 466 ppm
               despite the thin amount  of data between 443 and
               466 ppm.  In addition,  the NOx model does not
               exhibit unexpected  or inappropriate behavior at
               sulfur levels of 450 ppm.

     Based on the foregoing discussion, the final data range
limits enclosing the bulk of the data are summarized in Table IV-
42 .

                           Table IV-42
                 Data Core for the Complex Model
Pollutant
Aromatic s
Benzene
E200
E300
Olef ins
Oxygen
RVP
Sulfur
Units
vol%
vol%
vol%
vol%
vol%
wt%
psi
ppm
Limits of Data Core
Low end
18
0.4
33
72
1
0.0
7
10
High end
46
1.8
66
94
19
3.4
10
450
                               189

-------
2.    Valid Ranges

     The valid range for the Complex Model is here defined as
that range of fuel parameters over which the Complex Model (and
any applicable extrapolations as described in Section D.5) can be
used to evaluate the emission performances of fuels.  The Complex
Model can be used outside the valid range only if the model is
augmented through testing.  The valid range is not meant to place
absolute fuel parameter restrictions on either the reformulated
gasoline (RFG) or anti-dumping programs, but rather on the use of
the Complex Model within those programs.

     Specification of the valid range is a means of overcoming
some of the limitations of the Complex Model database.  As
described in the previous section, the database from which the
Complex Model is derived does not cover all possible fuel blends
that might be encountered in the RFG and anti-dumping programs.
The limits of the data core (as defined in Section D.I)  reflect
the limitations of the database and therefore the limitations of
the empirically-based Complex Model regression equations.  In an
effort to mitigate the disparity between the limitations of the
database and the needs of the RFG and anti-dumping programs,  the
Agency has extrapolated the Complex Model regression equations to
cover as many fuel blends as possible.  Even with extrapolation,
however, the Complex Model must be limited to some subset of the
fuels space to maintain the accuracy of its emissions performance
estimates.   These limitations on the valid range of the Complex
Model are discussed in this section.

     There are a number of factors which must be considered in
determining the valid range:

•    The purpose of the valid range

     Specification of the valid range is intended to make the
     Complex Model as flexible and useful as possible without
     sacrificing accuracy and reliability.  Insofar as the
     regression equations and their associated extrapolations are
     trustworthy, they should be applied to as wide a range of
     fuel parameters as possible.

•    The needs of the program

     The intent in developing the Complex Model was and is to
     provide refiners flexibility in meeting the requirements of
     the Clean Air Act.  Any restrictions in the valid range
     limit that flexibility.

     The valid range for the Complex Model is specific to the
     fuels program under which the Complex Model will be used,


                               190

-------
resulting in two separate valid range sets:  one for the RFC
program, and a second for the anti-dumping program.  There
are two reasons for delineating two separate valid range
sets:

1)    EPA believes it appropriate to permit the Complex
     Model to evaluate typical 1990 (non-reformulated)
     gasolines in order to determine compliance with
     anti-dumping regulations.  However,  EPA expects
     reformulated gasolines to fall into more
     restricted fuel parameter ranges than conventional
     gasolines.  Thus to fulfill the needs of each
     program, the valid range for the anti-dumping
     program should be less restrictive than the valid
     range for the RFC program.

2)    The accuracy of the Complex Model is less
     important for the anti-dumping program than it is
     for the reformulated gasoline program.  EPA
     believes that greater confidence is needed in
     estimating emission reductions from the use of
     reformulated gasolines because of the program's
     primary mission of achieving emission reductions
     in the worst ozone non-attainment areas.  The
     anti-dumping program for conventional gasolines
     only requires that future gasolines produce
     emissions which are no worse than those of
     gasolines produced in 1990.  Thus the
     environmental impact of inaccuracies in the
     model's emission estimates is more severe for the
     RFG program than for the anti-dumping program.

The behavior of the Complex Model

When the Complex Model predicts emission effects for a given
fuel parameter which seem questionable, linearly
extrapolating beyond the data core is less reliable,
indicating that the valid range for that fuel parameter
should approach coincidence with the data core.

Even when the model's predictions seem reasonable, the
mathematical form by which the model incorporates emission
effects may only be appropriate within the data core.  This
fact is of greatest concern when examining the location of
extrema (maximums and minimums).  Extrema produce a
"turnover" effect that is an artifact of using second-order
polynomials during the regressions, when the underlying
phenomenon may obey a decaying exponential function or some
other function which does not produce "turnover" effects.
Thus the model's emission performance estimates for fuels
with parameter levels that are outside the data core and are
                          191

-------
far from an extrema point may be much higher or lower than
the true effect.  Depending on the location of extrema,
placing the valid range limits too far from the data core
could result in emission performance estimates that differ
excessively excessively from the true values.

In some circumstances,  it may be inappropriate to linearly
extrapolate first-order relationships beyond the limits of
the data core.  A purely linear relationship may overstate
or understate the impact of fuel parameter modifications on
emissions,  particularly for fuel parameter levels which are
well beyond the limits of the data core,  since the linear
relationship is only an approximation of the underlying
phenonomenon being modeled.  Thus some linear relationships
may be most appropriately restricted to the limits of the
dat core.

The limits of the available data

The further one travels from the data core, the less
reliable are the emission performance estimates provided by
the model,  regardless of extrapolation.  Thus the size of
the data core will affect the size of the valid range.

Specific valid range proposals from sources outside EPA

In response to the July 15, 1993 package submittal to the
docket, a number of commenters provided EPA with specific
proposals for the valid range for both the RFC and anti-
dumping programs.  These valid range proposals provided EPA
with a valuable benchmark,  and are shown in Tables IV-43 and
IV-44.  Note that "N/A" indicates that a specific proposal
was not made, while "none"  indicates that the proposal was
for no valid range limit whatsoever.
                          192

-------
                   Table IV-43
Proposals for Reformulated Gasoline Valid Ranges
              for the Complex Model
Low end of range

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
EPA
0
50
6
33
76
15
2
0.2
API
0
0
6
30
70
0
0
0
AAMA
0
0
6.7
40
80
15
0
0
Penzoil
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
ARCO P
0
0
N/A
30
70
0
0
0
Exxon
N/A
10
6
N/A
N/A
10
2
0
Mobil
0
10
6
33
76
10
0
0
High end of range

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
EPA
3.7
500
10
65
94
47
20
2
API
3.7
500
15
70
100
55
25
2
AAMA
3.7
500
10
65
95
40
20
1
Penzoil
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
ARCO P
3.7
500
N/A
70
100
55
25
2
Exxon
N/A
600
15
N/A
N/A
55
30
2
Mobil
3.7
500
15
72
94
47
20
0
                       193

-------
                           Table IV-44
        Proposals for Conventional Gasoline Valid Ranges
                      for the Complex Model
Low end of range

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
EPA
0
0
6
30
70
10
0
0
API
0
0
6
30
70
0
0
0
AAMA
0
0
6.7
35
75
10
0
0
Penzoil
0
0
6
30
70
10
0
0
ARCO P
none
none
none
none
none
none
none
none
Exxon
none
none
none
none
none
none
none
none
Mobil
0
0
6
25
70
10
0
0
High end of range

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
EPA
3.7
1000
10
70
95
50
25
4.9
API
3.7
1000
15
70
100
60
30
4.9
AAMA
3.7
1000
10
70
95
40
25
2
Penzoil
3.7
1000
12
70
95
55
45
5
ARCO P
none
none
none
none
none
none
none
none
Exxon
none
none
none
none
none
none
none
none
Mobil
3.7
1000
15
72
95
50
25
4.9
     The Agency used all of the above criteria in specifying the
valid range for the Complex Model.  However, the Agency relied
most heavily on the behavior of the Complex Model, the limits of
the data core, and the valid range proposals from outside
parties, particularly when such proposals were supported by
available information.  These three factors include elements from
all the other criteria, and have the most direct and definable
impact on the valid range.  The needs of each program are
reflected in the valid range proposals in that commenters have a
unique perspective on the business ramifications of valid range
limitations.  The valid range proposals reflect the purpose of
specifying the valid range to provide the greatest flexibility to
suit private interests.  The Agency therefore seriously
considered each of the valid range proposals and weighed them
against the available data and observed behavior of the model
before specifying the final valid range.  The final values are
shown in Table IV-45.

     EPA defined the oxygen limits so as to encompass all
oxygenated gasolines which meet EPA's "substantially similar"
requirements, including a small blending tolerance.  Similarly,
the E200, lower E300, and upper aromatics limits were set near to
                               194

-------
those of the data core while making allowances for these
parameters' measurement and blending tolerances.  The upper limit
for E300 was not constrained since high E300 values appear to
result in reduced emissions.  However,  EPA restricted its linear
extrapolation of E300 effects to 95 volume percent; fuels with
higher E300 values must be evaluated as if their E300 levels were
equal to 95 volume percent.  In effect, this approach gives such
fuels no additional benefits for E300 increases beyond 95 volume
percent since EPA is not confident in its ability to quantify
these benefits.  Nonetheless, EPA expects such increases to
result in lower emissions, so EPA chose not to restrict the upper
E300 limit of the model.

     Aromatics reductions result in NOx reductions, with the
benefits increasing as aromatics levels decrease.  As a result,
EPA's extrapolation of aromatics' effects likely underestimates
the benefits of aromatics reductions below 18 vol%.  Since EPA
believes its model is unlikely to overestimate the benefits of
very low aromatics levels, EPA set a relatively unconstraining
lower aromatics limit of 10 volume percent when linearly
extrapolating its models.  In addition, EPA decided to allow
fuels with aromatics levels below 10 volume percent to be
certified using the Complex Model, but such fuels' emission
performance must be evaluated as if their aromatics levels were
at 10 volume percent.  This approach provides refiners with
greater flexibility without imposing significant risks on the
environment.  Olefin limits were based on similar reasoning:  EPA
considers the risk that its model underestimates the NOx
increases associated with olefin levels beyond 19 vol% to be
minimal.

     EPA chose not to extend the upper limit of the valid range
for sulfur beyond 500 ppm for the reformulated gasoline program
because the NOx model had an emissions maximum at approximately
550 ppm sulfur.  This maximum appears to be an artifact of the
form of the model (see Section D.4 for further discussion).   The
available data indicates that as sulfur levels increase,
emissions increase but at a decreasing rate.  Hence the straight-
line extrapolation used by EPA would be expected to deviate
significantly from the underlying "true" effect of sulfur as
sulfur levels increase to levels substantially above the upper
limit of the data core.  This deviation is particularly important
for the RFC program, which requires fuels to meet absolute
emission performance standards.  The anti-dumping program,
however, requires refiners to meet emission standards relative to
their 1990 baseline emissions; as a result, the accuracy of the
extrapolation is less critical.  EPA has set a less restrictive
upper limit to the valid range for anti-dumping since it
considers the gain in flexibility to outweigh the possible losses
due to less precise extrapolation.
                               195

-------
     The upper RVP limit was set so as to not exclude any summer
gasolines which might be sold in 1995 or later.  The lower limit
was set so as to encompass the RVP levels that EPA expects to
characterize Phase II reformulated gasolines, including the
associated measurement tolerance.  EPA decided against setting a
less restrictive lower limit based on concerns expressed by some
commenters that such gasolines might produce driveability
problems.  Poor driveability tends to result in increased
emissions, and EPA has no data which demonstrate emission
benefits of lowering RVP levels below 7 psi.

     The only pollutant affected by fuel benzene levels is
benzene itself.  The upper limit for RFC is largely irrelevant,
since benzene is limited to a maximum of 1.3 vol% under averaging
and 1 vol% under per-gallon compliance.  Gasolines containing 5
vol% or more benzene must be labeled as hazardous materials under
existing OSHA requirements.  EPA believes that this labeling
requirement has prevented marketing of such fuels in the past and
will continue to do so in the future.  For this reason, EPA
believes that the 4.9 vol% limit on conventional gasolines
encompasses all such fuels that are likely to be produced.

                           Table IV-45
             Final  Valid  Ranges  for  the  Complex Model

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
Reformulated Gasoline
Low end
0
0
6.4
30
70
10
0
0
High end
3.7
500
10
70
95
50
25
2
Conventional Gasoline
Low end
0
0
6.4
30
70
0
0
0
High end
3.7
1000
11
70
100
55
30
4.9
3.    Extrapolation Methodology

     The regression equations that comprise the Complex Model are
empirical in nature.  Thus the strictest possible application of
these equations would limit their use to the range of fuel
parameters specified by the data core.  As discussed in Section
D.2, however,  the valid range of the Complex Model has been
extended beyond these limits.  To use of the Complex Model
outside the data core,  EPA has extrapolated the regression
equations in such a manner as to maintain the integrity and
accuracy of the relationships between fuel properties and
emissions.  This section provides the mathematical basis for
                               196

-------
model extrapolation, while Section D.5 applies this methodology
to each of the pollutant models.

     The choice of extrapolation methodology applied to a given
regression equation depends on one's confidence in the accuracy
of the underlying regression equation.  The regression techniques
considered by EPA include:

•    Regression curve approach

     This approach uses the regression curve as-is for all areas
     of the fuels space.  It is easy to implement and requires no
     changes to the derived regression equations.  However,  it
     assumes that the regression equation accurately captures the
     underlying relationship between fuel properties and
     emissions.  It also assumes that trends seen within the data
     core continue at levels beyond the data core.

•    Flat-line approach

     This approach uses the regression curve only within the
     limits of the data.  It holds fuel parameters which fall
     outside the data core at the nearest limit of the data core
     when evaluating fuel performance by the Complex Model.   In
     graphical form, the slope of the emission performance versus
     fuel parameter level curve discontinuously becomes zero
     (i.e., flat) for fuel parameter levels equal to or beyond
     the nearest limit of the data core.

     This approach is attractive because it is straightforward to
     implement.  It is conservative from a statistical standpoint
     since it assumes that nothing can be inferred from the
     regression equations about areas of the fuels space not
     covered by the data core.  However, refiners would be given
     no incentive at the margin to make high-emitting fuels
     cleaner if they fall outside the data core.  In addition,
     this approach ignores the results of studies not included in
     the Complex Model database which indicate that the emission
     performance versus fuel parameter level curve has a non-zero
     slope outside the data core.

•    Straight-line approach

     This approach uses the first derivative of the Complex Model
     equations to extrapolate those equations.  It assumes that
     the relationship between emissions performance and fuel
     parameter level at parameter levels beyond the data core can
     be approximated accurately by the slope of the emissions
     performance versus fuel parameter curve  (in multi-
     dimensional space) at the limits of the data core.
                               197

-------
     This approach preserves the non-zero fuel parameter remission
     relationships observed within the data core.  It avoids
     inferring second-order effects for fuels outside the data
     core.  However, this approach can be complex to implement
     due to the need to take derivatives.  In addition,  the slope
     observed at the limits of the data core may differ from the
     actual slope outside the data core,  though this problem is
     mitigated to some degree since the difference is likely to
     remain small for fuel parameter values that are relatively
     close to the limits of the data core.

     In addition to the choice of extrapolation approach,  the
choice of the point from which to start the extrapolation also
must be determined before specific extrapolations can be applied
to the Complex Model.  Ideally, this point would coincide with
the limits of the data core.  However, the mathematical form of
the exhaust complex model is limited to first- and second-order
functions of fuel parameters.  The actual relationship between
fuel properties and emission levels is unlikely to take this
form.  A full mathematical model of the underlying phenomena or
the data included in the Complex Model database may require
decaying or growing exponentials, third-order polynomials,  square
root relationships, or even more complex functions.  The Complex
Model approximates the actual relationships between fuel
properties and emissions using first- and second-order
polynomials; such approximations can result in extrema  (emission
maximums or minimums) which are artifacts of the form of the
exhaust Complex Model equations and which do not reflect the
underlying data used to develop the model.  These inappropriate
extrema may occur within or outside of the data core.  The
extrapolation approaches described above can be used to correct
such inappropriate extrema by selecting an appropriate starting
point for the extrapolation, such as the extrema point itself.
The location of extrema in the Complex Model will be discussed in
Section D.4, while the ramifications of those extrema will be
addressed in Section D.5.

     The regression curve and flat-line approaches to
extrapolation are mathematically straightforward to derive and
apply.  The remainder of this section develops and presents the
mathematical basis for the straight-line approach to
extrapolation.  This approach begins with the performance
calculation procedure described in Section C but includes the
following additional steps:

     1)   Determine which fuel parameters for the target
          fuel fall within the extrapolation range but
          outside the "allowable range,"  which is the range
          within which the unmodified regression equations
          can be used.  The extrapolation range and
          allowable range for all fuel parameters are
          specified in Section D.5.

                               198

-------
     2)    Determine the fuel parameter values for an "edge
          target fuel" in which the fuel parameters that
          would naturally fall within the extrapolation
          range are held at the limits of the allowable
          range.  Target fuel parameters that are within the
          allowable range should remain at their target fuel
          levels when defining the edge target fuel.

     3)    Evaluate the performance of the edge target fuel.
          This step yields a performance estimate at the
          edge of the allowable range, or an "edge
          performance" symbolized by Y(et),  where "et"
          refers to "edge target fuel."

     4)    Determine the multi-dimensional slope of the
          emissions versus fuel parameter curve at the point
          in the multi-dimensional fuels space defined by
          the edge target fuel's composition.  This step is
          accomplished by calculating the first derivative
          of the exhaust regression equation(s).

     5)    Compute the additional performance benefit or
          detriment due to the fuel parameter(s)  that are
          beyond the allowable range.  This step yields an
          "extrapolation performance factor" symbolized by
          AY which is calculated by multiplying the
          difference between the target fuel parameter level
          and the nearest limit of the allowable range by
          the slope calculated in step (4).

     6)    Add the edge performance to the extrapolation
          performance factor to obtain the oveerall
          oerformance of the original target fuel.

These six steps can be mathematically summarized as

     (D.I)     Y(t) = Y(et) + AY

Determining the fuel parameter values for the edge target fuel
and calculating performance estimates for the edge target fuel
need no elaboration beyond that given in the steps above.
Calculating AY,  however,  is considerably more complex because it
requires determining the first derivative of the exhaust model
for each pollutant.

     As described in Section D.5, the entire non-exhaust model is
extrapolated by way of the regression curve approach, i.e., the
non-exhaust emissions for any fuel within the valid range can be
evaluated directly from the non-exhaust equations.   Thus only the
exhaust model equations are candidates for straight-line
extrapolation.  And since the calculations in the exhaust model


                               199

-------
are first carried out in percentage change terms and must
subsequently be converted into gram/mile estimates, it is most
appropriate to use the percentage-change performance as the
dependent variable in determining the first derivative for the
exhaust models.  As described in Section C above, the exhaust
pollutant performance of a target fuel is calculated from
Equation (C.7):

     (D.2)      Y(t) = £i{Wi *  [exp(fi(t))/exp(fi(b))  -  1]  *  100}

where Y is the fleet average percent change in any exhaust
pollutant,  the wi  are the vehicle group weighting factors,  t
refers to the target fuel, and b refers to the baseline fuel.
This is the equation from which the first derivative, and thus
the extrapolation performance factor AY,  will be derived.

     In order to determine the extrapolation performance factor,
the first derivative of Equation  (C.7) must be determined for the
edge target fuel as described in step  (4).  Since the derivative
of a sum is equivalent to the sum of the derivatives, the
derivative of each vehicle group term in Equation  (D.2) can be
evaluated separately.  Thus Equation  (D.2) is first separated
into its vehicle group-specific terms:

     (D.3)      Y(et)i =  w± *  [exp(fi(et))/exp(fi(b))  -  1]  *  100

Equation (D.3) is shown as being evaluated for the edge target
fuel, consistent with step 3) above.  The total derivative of
Equation (D.3) is derived from the partial derivatives of
Equation (D.3) with respect to each of the fuel parameters.
Mathematically, this total derivative can be written in its
general form as

     (D.4)      dYi =  EJOYi/dPjJdPj

where the Pj  are  the  possible fuel parameters in the Complex
Model equations for the pollutant of interest.  As before, the
subscript i refers to the vehicle group, of which there are two
in the final Complex Model: one for normal emitting vehicles, and
a second for higher emitting vehicles.  Note that linear
extrapolation of a first-order equation is identical to use of
the unmodified original equation and is therefore equivalent to
the the regression curve extrapolation approach.  Thus the
straight line extrapolation approach need not be applied to fuel
parameters that are only found in first-order terms.

     The total derivative of Equation  (D.2) can be written
generally as

     (D.5)      dY = Xi(Wi  *dY±)  = Xi  [w± * I
                               200

-------
The general Equation  (D.4) can be  applied specifically to
Equation  (D.3) and the results weighted  by the wi weighting
factors and then summed to obtain  the  Complex Model-specific
results of Equation  (D.5).   In other words,  the solution to
Equation  (D.4) is sufficient  to  solve  Equation (D.5).

In order to best illustrate  the  evaluation of Equation (D.4),  a
hypothetical second-order function f can be created.   Eliminating
the vehicle group distinction indicated  by the subscript i and
instead making the vehicle group distinction implicit,  this
function can be written as

      (D.6)     f(A,B) = e1A+e2B+e3A2+e4B2+e5AB

In the above equation, A and  B are hypothetical uncentered fuel
parameters, and the coefficients e± are determined  through
statistical regressions.  Note that all  possible combinations of
linear, squared, and  interactive terms have been included in this
hypothetical function.  Thus  the vehicle group-specific equation
giving the performance of a  target fuel  as a function  of the fuel
parameters A and B is

                    exp(e1At+e2Bt+e:At2+e1Bt2+e;AtBt)
      (D.7)  Y(t) = lexp(eiAb+e2Bb+e3Ab2+e4Bb2+e5AbBb)  "     I * 100

where Y(t) is the performance of the target fuel for any
pollutant, At is the value of fuel parameter A  for  the  target
fuel, Bb is the value of fuel parameter B  for the baseline fuel,
and so on.  To determine the  total first derivative of Y,  partial
derivatives of Y with respect to each  of the fuel parameters must
be derived analytically and  summed according to the following
equation:

      (D.8)     dY =  OY/dA)dA +   OY/dB)dB

dY is called the differential of Y.  Since the total derivative
will be used to determine the slope of the curve at the edge
target fuel, each of  the partial derivatives must be evaluated at
the edge target fuel:

      (D.9)     dY =  (8Y/8A)ledge dA + (dY/dB)ledge dB

Since the Complex Model is at most a second-order model (in
logarithmic space), the first derivative must be first order.
This fact implies that the total differentials for  Y,  A,  and B in
Equation  (D.9) can be written as finite  deviations  from the edge
target fuel:

dY = AY = Y(t)  - Y(et)
                                201

-------
dA = AA = At  - Aet   if At < low end of allowable range or
                       At > high end of allowable range

        =0         if At > low end of allowable range and
                       At < high end of allowable range

dB = AB = Bt  - Bet   if Bt < low end of allowable range or
                       Bt > high end of allowable range

        =0         if Bt > low end of allowable range and
                       Bt < high end of allowable range

The conditional  statements associated  with  each fuel  parameter
differential quantify the  following conditional statement:

     If the value of a fuel parameter  for a given  target
     fuel lies within the  allowable range,  then do  not  use
     the first derivative  of the model to extrapolate the
     function Y  with respect to that  fuel parameter;
     otherwise,   linearly extrapolate  from the  edge  target
     fuel to the fuel parameter value  of the target fuel.

Thus a given fuel parameter differential will  be zero if that
fuel parameter lies within the allowable range,  but will be
nonzero if it lies outside the allowable range.  Extrapolation
occurs only for  fuel parameters which  lie beyond the  allowable
range.

     The partial derivatives of pollutant Y with respect to each
of the two fuel  parameters A and B are shown below.   Note  that
f(b) can be considered a constant since the baseline  fuel  does
not change.

dY/dA      =  [exp(f (t) )/exp(f (b) ) ] * 100 * df(t)/8A

          =  [exp(f (t))/exp(f (b))] * 100 *  (e,  +  2*e3At +  e5Bt)

dY/dB      =  [exp(f (t) )/exp(f (b) ) ] * 100 * df(t)/dB

          =  [exp(f (t))/exp(f (b))] * 100 *  (e2  +  2*e4Bt +  e5At)

The total derivative for Y can now be  assembled for this
hypothetical regression equation, and  the value of  Y  for a  given
target fuel calculated as  a function of the fuel parameters A and
B.  When Equation  (D.9) is evaluated with the  differentials and
partial derivatives shown  above, the  specific  form  of Equation
(D.I)  is produced:

Y(t) =    Y(et)   +
          [exp(f (t))/exp(f (b))]*100*(e1 + 2*e3At+e5Bt)  *  (At -  Aet) +
          [exp(f (t))/exp(f (b))]*100*(e2 + 2*e4Bt+e5At)  *  (Bt -  Bet)
(D.10)

                               202

-------
And therefore, from Equation  (D.I),

A¥ =  [exp(f (t))/exp(f (b))]*100*(e1 + 2*e3At+e5Bt) *  (At - Aet)  +
      [exp(f (t))/exp(f (b))]*100*(e2 + 2*e4Bt+e5At) *  (Bt - Bet)
(D.ll)

Note that the conditional statements described above still apply
to the fuel parameter differentials  (given as differences) in
Equation (D.10).

     Equation  (D.10) is a vehicle group-specific equation, and
therefore represents a solution to Equation  (D.I) only for a
single vehicle group; Equations (D.10) and  (D.ll) actually give
the solutions for Y±(t)  and AYI; respectively.  To obtain the
fleet average solution to Equation (D.I), the vehicle group-
specific solutions must be weight-averaged with the vehicle group
weighting factors wi according to  Equation (D.5).   Thus dYi  (or
AYJ has been  determined only  for  a single vehicle  group  in  this
example, and must be combined with the solutions for dYi for the
other vehicle group in order to determine the overall exhaust
emission performance of the target fuel.

4.   Extrema

     As discussed in the previous section, the second-order
polynomial form of the Complex Model can  introduce  inappropriate
extrema into the model.  In fact,  all second order  equations
contain extrema points at which the direction of the emission
effect changes (from increasing to decreasing or vice-versa).
The Agency has considered a number of approaches to mitigate the
effects of inappropriate extrema on predicted emissions using the
unmodified Complex Model.  These approaches and the Agency's
concerns with each are reviewed below:

     1.    The Complex Model could be redeveloped from the
          original database using growing or decaying exponential
          functions instead of simple second-order  polynomials.

          This approach replaces one assumed functional form for
          the underlying combustion and catalyst chemistry with a
          different assumed functional form.  Replacing only
          those second-order relationships which would be better
          represented with a decaying exponential would require a
          far deeper understanding of combustion and catalyst
          chemistry than is currently available.  This approach
          also raises the possibility that other non-polynomial
          functions should be considered, would compromise the
          consistency in model form between pollutant-specific
          models, and could make the use  of the random balance
          method to consolidate the models for normal emitter
          technology groups impossible.
                               203

-------
2.    Any second-order terms which result in questionable
     behavior could be pre-excluded from the Complex Model,
     which would then be re-estimated.

     EPA rejected this approach after careful evaluation,  as
     indicated by the models included in the July 14,  1993
     docket submittals.  This approach relies on strictly
     first-order effects,  which can result in behaviors that
     are just as questionable as the second-order effects.
     Also,  linear extrapolation beyond the limits of the
     data for the affected terms becomes unjustifiable since
     the model's reliability and accuracy would decrease
     rapidly as one moved beyond the data core.  Further,
     EPA's efforts to build such models resulted in poorer
     predictability.  Finally,  EPA found that because of the
     covariation between fuel parameters, pre-excluding one
     second-order term resulted in the introduction of a
     second-order term for a covarying parameter.  For
     example, when the ARO2  term was  pre-excluded,  the
     resulting model included an E3002  term that  was just  as
     troublesome as the original ARO2 term.

3.    One or more interactive terms could be forced into the
     model in an effort to replace the squared term causing
     the inappropriate behavior with better-behaved
     interactive terms.

     This approach assumes that the statistical derivation
     of the Complex Model has assigned an effect to a
     squared term which should instead be assigned to an
     interactive term.  All attempts by the Agency to
     implement this approach resulted in a model which
     retained the troublesome squared term or which suffered
     from other behavioral shortcomings.

4.    Flat-line extrapolations could be imposed from the
     extrema points.  This approach,  which was supported by
     a number of commenters, is illustrated in Figure IV-1,
     which also shows a decaying exponential function.
     Statistical and chemical considerations suggest that
     some of the second-order terms in the complex model
     result from phenomena which exhibit such underlying
     behavior.
                          204

-------
                 Figure IV-1:   Mitigating Extrema
                                   Flat-line extrapolation
        Decaying exponential
                              Second-order polynomial
                            Fue I param eter
          This approach required the  fewest  assumptions  regarding
          the form of the regression  equation,  and it  resulted in
          models which behaved in a manner consistent  with the
          available data.  In addition,  it most closely
          approximated a decaying exponential  effect without
          compromising the form of the model.

The Agency determined that flat-line  extrapolation was the most
appropriate method for mitigating questionable effects caused by
extrema.

     Applying a flat-line extrapolation  from an extremum point is
equivalent to linearly extrapolating  the model from the  same
point.  At the extremum, the slope calculated  by the linear
extrapolation method is zero  (in fact, a zero  first derivative is
a necessary condition for an extremum) and is  thus identical to
the slope imposed by a flat-line extrapolation from that point.

     The use of flat-line extrapolation  implies that the
previously stated conditions under which linear extrapolation is
to be used must be supplemented.  The points at which  linear
extrapolations are to begin  (i.e., the limits  of the allowable
range) are defined by one of the following two considerations:

     1) The limits of the data core.
     2) The location of extrema.
                               205

-------
The data core is the default criterion since it defines the range
of fuel parameters within which the regression equations are most
trustworthy.  Insofar as the data core does not reach the limits
of the valid range, linear extrapolation must be used to extend
the regression equations without placing an unwarranted amount of
trust in their curvature.  If,  however,  an extremum point is
located within the limits of the data core, the model may behave
inappropriately.  In such cases, the extremum point replaces the
nearest data core limit as the limit of the allowable range, and
linear extrapolation begins at that point instead of at the end
of the data core.

     Since the Complex Model regression equations are
multivariate, separate extrema can exist for each fuel parameter.
Thus the partial derivative of the regression equation with
respect to a given fuel parameter must be calculated, set equal
to zero, and solved for the fuel parameter of interest to
determine the extremum location.  When this is done for the
Complex Model, the extrema locations shown in Tables IV-46 and
IV-47 are identified.  Note that "N/A" indicates that either a
strictly linear relationship exists between the fuel parameter
and pollutant in question, or there is no relationship
whatsoever.

                           Table IV-46
         Phase I Extrema Locations for the Complex Model

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
MTBE
ETBE
Ethanol
TAME
VOC
N/A
N/A
N/A
65.8
80.3-101.8
N/A
N/A
N/A
N/A
N/A
N/A
N/A
NOx
N/A
554
N/A
N/A
N/A
36.2
3 .77
N/A
N/A
N/A
N/A
N/A
Toxics*
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
                               206

-------
                           Table IV-47
         Phase  II  Extrema Locations for the Complex Model

Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
MTBE
ETBE
Ethanol
TAME
VOC
N/A
N/A
N/A
65.5
79.7-100.9
N/A
N/A
N/A
N/A
N/A
N/A
N/A
NOx
N/A
578
N/A
N/A
N/A
36.8
3 .77
N/A
N/A
N/A
N/A
N/A
Toxics*
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
  All four toxics models are linear, and therefore contain no extrema
5.    Extrapolation of the Final Complex Model

     The linear extrapolation methodology for the Complex Model
equations as described above requires the pollutant-specific
forms of Equation  (D.I) and the specific points at which
extrapolation is to begin.  Calculation of the first derivative
and application of flat-line extrapolation from extrema are
required only for the exhaust VOC and exhaust NOx models.  The
exhaust toxics models are strictly linear, and they and the non-
exhaust VOC and benzene models can be extrapolated via the
regression curve approach (see Section D.3 for a description of
the various extrapolation methodology options).  In this section,
the definition of the edge target fuel will be reiterated, the
points at which extrapolation is to begin (which define the
allowable ranges) will be presented, and then the pollutant-
specific forms of Equation (D.I) will be derived and presented
for both exhaust VOC and NOx.

     As described in Section D.3, linear extrapolation beyond the
allowable range requires the definition and evaluation of an edge
target fuel.  It is at the position of the edge target fuel that
both the edge target fuel performance and the (multi-dimensional)
slope of the curve are calculated.  The slope of the curve is
used to determine the extrapolation performance factor, and
Equation (D.I) is then used to determine the extrapolated
performance of the target fuel.  The edge target fuel represents
the position in the multi-dimensional fuels space that is closest
to the position of the target fuel without going outside the
limits of the allowable range defined below.  Since the allowable
range for a given fuel parameter is pollutant-specific, the edge
target fuel must also be defined separately for each pollutant.
The edge target fuel can therefore be defined as follows:
                               207

-------
     All fuel parameters of the edge target fuel are
     identical to those of the target fuel with the
     following exception: if a given target fuel parameter
     is outside of the allowable range,  then hold the
     equivalent edge target fuel parameter to the limit of
     the allowable range which is closest in value to the
     target fuel value.

Thus if the target fuel value of aromatics is 52 vol% and the
allowable range is defined as 18 to 46 vol%,  then the edge target
fuel aromatics content should be 46 vol%, since 46 vol% is closer
to 52 vol% than is 18 vol%.

     The data core and the location of extrema points were
specified above to delineate the allowable range, which is the
range of fuel parameters within which the Complex Model's exhaust
emission equations can be used without modification and outside
of which linear extrapolation must be employed.  The general
guidelines used by EPA to establish the allowable ranges of fuel
parameters which are included in second-order  (either squared or
interactive)  terms can be stated as follows:

     If an extremum point for a given fuel parameter is
     located within the data core limits for that fuel
     parameter, it indicates inappropriate model behavior.
     Therefore, linear extrapolation must begin at the
     extreme point and proceed in the direction deemed most
     appropriate.  If an extreme point is not located within
     the data core, linear extrapolation must begin at the
     limits of the data core.

The appropriate direction of the linear extrapolation from an
extremum point depends primarily on the following two factors:

     1)   The position of the extremum point relative to the
          limits of the data core.

     2)   The underlying combustion and catalyst chemistry, if
          known.

     2)   The emission trends identified in specific studies or
          in the Complex Model database.

Linear extrapolation of a given fuel parameter from the low end
of the data core proceeds downward to lower parameter values,
while linear extrapolation from the high end of the data core
proceeds upwards to higher parameter values.   Any regression
equation containing fuel parameters that are strictly linear need
not be extrapolated using the straight-line,  first derivative
                               208

-------
technique  since such extrapolation is equivalent to  using the
unmodified equation.46

     Based on the foregoing discussion,  EPA has defined the
allowable  range for all second-order fuel parameters included in
the Complex Model as shown in Table IV-48.

                            Table  IV-48
        Allowable Parameter Ranges  for Use of  the Complex
                     Model  Regression  Equations
Fuel
parameter
Sulfur
E200
E300
Aromatic s
Olef ins
Pollutant
NOx
VOC
VOC
VOC
NOx
NOx
Phase I
Low
end
10
33
72
18
18
3 .77
High
end
450
65.83
vary*
46
36.2
19
Phase II
Low
end
10
33
72
18
18
3 .77
High
end
450
65.52
vary**
46
36.8
19
           The high end of the allowable range (in vol%) equals 79.75 +
           0.3846*ARO, where ARO is  the aromatics content of the target fuel,
           or 94 vol%, whichever is  lower.


           The high end of the allowable range (in vol%) equals 80.32 +
           0.3903*ARO, where ARO is  the aromatics content of the target fuel,
           or 94 vol%, whichever is  lower.
Only the  allowable ranges for  those fuel parameters  present as
second-order terms are shown in  Table IV-48; for all other fuel
parameters,  the Complex Model  equations can be used  without
modification out to the limits of  the valid range.   Thus the
"allowable  range" differs from the "valid range" only for those
fuel parameters with second-order  effects on a given pollutant.

     The  high and low values in  Table IV-48 bracket  the range of
values for  a given fuel parameter  within which the exhaust
regression  equations can be used without modification,  and
outside of  which straight-line linear extrapolation  must be
employed.   The values in Table IV-48 also indicate the points at
which the multi-dimensional slope  of the emissions versus fuel
         See discussion in Section D.I regarding the effect of an exponential

function on the behavior of polynomials.

                                 209

-------
parameter relationship must be evaluated when linear
extrapolation is required.  For fuel parameters that contain an
extremum point somewhere within the data core, the location of
that extremum point defined one end of the allowable range.  Note
that the allowable ranges given in Table IV-48 apply only to the
exhaust VOC and NOx portions of the Complex Model; the exhaust
toxics models and the nonexhaust VOC and benzene models can be
used without modification throughout all areas of the fuels
space, based on the regression curve approach to extrapolation
described in Section D.3.

     The high end of the allowable range for the effect of E300
on VOC is variable due to the ARO*E300 term in the exhaust VOC
model.  Since the upper and lower limits of the data core for
E300 are 72 vol% and 94 vol%, respectively, an extremum point
between these values must, by the conditions described above,
define one end of the allowable range.  EPA derived a
relationship defining these extrema by taking the first
derivative of the exhaust VOC equations:

      (D.12)    Phase I  E300 extremum =
                    80.32 vol% + 0.3903 x aromatics vol%

      (D.13)    Phase II E300 extremum =
                    79.75 vol% + 0.3846 x aromatics vol%

As per the discussion of extrema in Section D.4, Equations (D.12)
and (D.13)  indicate that the E300 extremum point can be greater
than or less than 94 vol% depending on the target fuel aromatics
content.  Thus the upper end of the allowable range for the
effect of E300 on exhaust VOC is conditional:

     If the E300 extremum as calculated from Equation (D.12)
     or (D.13) is less than 94 vol%, then the E300 extremum
     becomes the upper limit of the allowable range for
     E300.   If the E300 extremum is greater than 94 vol%,
     then the upper end of the allowable range for E300 is
     94 vol%.

     For exhaust VOC, linear extrapolation must be used if E200,
E300,  or aromatics are outside of their allowable ranges.  To
obtain the exhaust VOC-specific form of Equation  (D.I),  the first
partial derivatives of the exhaust VOC equation must be taken
with respect to each of these three fuel parameters.  Since there
are two vehicle group exponential terms for all exhaust
regression equations, these partial derivatives must be derived
separately for each vehicle group.  From Section A.9, the final
exhaust VOC regression equation for Phase I is

   (D.14)  VOC(t) =  100% x 0.52 x  [exp(a(t))/exp(a(b))  - 1] +
                    100% x 0.48 x  [exp(b(t) )/exp(b(b) )  - 1]
                               210

-------
where the functions a and b are defined  as
a =
 -0.003641  x Oxygen)
 0.0289749    x RVP)
            x E300)
            x Olefins]
     [-0.068624
     [-0.002858
     [0.0004087
                 [0.0005219
                 [-0.014470
                 [0.0323712
                 [0.0001072
            x Sulfur)
            x E200)
            x Aromatics]
            x E2002)
b =
[-0.003626
[0.043295
x E3002)

  x Oxygen)
  x RVP)
     >0.062327 x E300)
     >0.002858    x Olefins;
[0.000408
                  x E3002
[-0.0003481  x Aromatics  x E300)

(-0.000054    x Sulfur)      +
(-0.013504 x E200)           +
(0.0282042  x Aromatics)    +
(0.000106     x E2002)       +
[-0.000287    x Aromatics x E300;
and a(t) refers to the evaluation of  function a for the target
fuel, b(b) refers to the evaluation of  function b for the
baseline fuel, and so on.  The  first  partial  derivatives with
respect to E200, E300, and aromatics  can  be derived separately
for normal emitting vehicles  (the first term  in Equation (D.14)
and higher emitting vehicles  (the second  term in Equation
(D.14)):
dVOC1/dE200 =
dVOCVdARO


dVOC2/dE200


dVOC2/dE300


dVOC2/dARO
           100%  x 0.52  x [exp(a(t) )/exp(a(b) ) ]  x
           [0.000214416 x E200 - 0.0144703]

           100%  x 0.52  x [exp(a(t) )/exp(a(b) ) ]  x
           [0.000817412 x E300 - 0.0686242 -  0.00034806 x
           ARO]

           100%  x 0.52  x [exp(a(t) )/exp(a(b) ) ]  x
           [-0.0003481  x E300 + 0.0323712]

           100%  x 0.48  x [exp(b(t) )/exp(b(b) ) ]  x
           [0.000212  x  E200 - 0.0135039]

           100%  x 0.48  x [exp(b(t) )/exp(b(b) ) ]  x
           [0.000816  x E300 - 0.0623274 - 0.000287 x ARO]

           100%  x 0.48  x [exp(b(t) )/exp(b(b) ) ]  x
           [-0.000287 x E300 + 0.0282042]
The subscript 1 indicates the normal  emitter  vehicle group,  while
the subscript 2 indicates the higher  emitter  vehicle group.   The
general extrapolation equation  for VOC,  per Equation (D.I),  is
written as follows:

     (D.15)    VOC(t) =
                     100% x 0.52 x
                     100% x 0.48 x
                     (8VOC1/8E200)iet
[exp (a (et)
[exp (b (et)
x AE200 +
x AE300 +
x AARO +
)/exp(a(b))
)/exp(b(b))
- 1] +
- 1] +
                               211

-------
                     (aVOC2/8E200)iet x AE200 +
                     (8VOC2/8E300)iet x AE300 +
                     (8VOC2/8ARO) iet  x AARO

Note that the first  two terms  are  now  evaluated for the edge
target fuel rather than the target fuel,  as  are the partial
derivatives.  The final extrapolation  equation for Phase I
exhaust VOC is therefore given by  Equation (D.16):

Phase I VOC extrapolation equation:

     (D.16)    VOC(t) =
               100%  x 0.52 x  [exp(a(et) )/exp(a(b) )  - 1]  +
               100%  x 0.48 x  [exp(b(et) )/exp(b(b) )  - 1]  +
               100%  x 0.52 x  [exp(a(et) )/exp(a(b) ) ]  x
                {AE200 x  [0.000214416   x E200iet - 0.0144703] +
                 AE300 x  [0.000817412   x E300iet - 0.0686242 -
                 0.00034806 x  ARQet]  +
                 AARO  x  [-0.000348061 x E300iet + 0.0323712]} +
               100%  x 0.48 x  [exp(b(et) )/exp(b(b) ) ]  x
                {AE200 x  [0.000212  x  E200iet - 0.0135039] +
                 AE300 x  [0.000816  x  E300iet - 0.0623274 -
                 0.000287 x ARQet]  +
                 AARO  x  [-0.000287  x  E300iet + 0.0282042]}

a =  (-0.003641 x Oxygen)     +  (0.0005219  x Sulfur)      +
     (0.0289749   x RVP)      +  (-0.014470  x E200)        +
     (-0.068624 x E300)       +  (0.0323712  x Aromatics)  +
     (-0.002858  x Olefins)   +  (0.0001072  x E2002)       +
     (0.0004087  x E3002)      +   (-0.0003481 x Aromatics  x  E300)

b =  (-0.003626    x  Oxygen)  +  (-0.000054    x Sulfur)      +
     (0.043295     x  RVP)     +  (-0.013504 x E200)           +
     (-0.062327 x E300)       +  (0.0282042  x Aromatics)  +
     (-0.002858    x  Olefins) +  (0.000106     x E2002)       +
     (0.000408     x  E3002)    +   (-0.000287    x Aromatics  x  E300)

The conditions under which the finite  differences  AE200, AE300,
and AARO are nonzero are identical to  the conditions under which
these fuel parameters require  extrapolation.   Thus for each of
these three fuel parameters which  is outside of the allowable
range as defined in  Table IV-48, the associated finite difference
will be equal to the difference between the  target fuel value and
the nearest limit of the allowable range.   However,  to simplify
the regulations, linear extrapolation  from extrema points was
replaced with flat-line extrapolations,  since for  extrema points
these two extrapolation methodologies  are equivalent.   The
allowable range conditions can be  translated into  the following
mathematical conditions:

Phase I Conditions for extrapolating exhaust VOC:
                                212

-------
     If  E200it  <  33,  then AE200  =  E200it  -  33
     Otherwise AE200 =  0

     If  E300it  <  72,  then AE300  =  E300it  -  72
     If  E300it  >  94 and 80.32  +  0.39.3 x ARQt > 94, then AE300 =
         E300it  -  94
     Otherwise AE300 =  0

     If  AROit <  18,  then AARO  =  AROit  -  18
     If  AROit >  46,  then AARO  =  ARQt  -  46
     Otherwise AARO = 0

These conditions guarantee  that the slope  of the VOC regression
curve at the position of  the  extrema points will be zero47 as
required by the definition  of extrema points.

     The Phase II extrapolated performance equation for VOC is
identical to equation D.16, except that  the vehicle group
weighting factors change:

Phase II VOC extrapolation  equation:
      (D.17)     VOC(t) =
                     100%  x 0.444 x  [exp(a(et
                     100%  x 0.556 x  [exp(b(et
                     100%  x 0.444 x  [exp(a(et
                       { [0.000214416 x E200iet -
                        [0.000817412 x E300iet -
                        0.00034806 x AROiet]  x
                        [-0.000348061 x E300iet
                     100%  x 0.556 x  [exp(b(et
                       { [0.000212 x E200iet -  0.
                        [0.000816 x E300iet -  0.
                        AROiet]  x AE300 +   AARO
                        [-0.000287 x E300iet + 0
                                          ))/0.0620896  -  1]  +
                                          ))/0.1037675  -  1]  +
                                          ))/0.0620896] x
                                           0.0144703]  x AE200 +
                                           0.0686242  -
                                          AE300  +   AARO x
                                          + 0.0323712] } +
                                          ))/0.1037675] x
                                          0135039]  x  AE200 +
                                          0623274  - 0.000287 x
                                           x
                                          .0282042] }
a =
[-0.003641 x Oxygen)
[0.0289749   x RVP)
[-0.068624  x E300)
[-0.002858  x Olefins;
[0.0004087  x
[0.0005219
[-0.014470
[0.0323712
[0.0001072
x Sulfur)
x E200)
x Aromatics]
x E2002)
                                  [-0.0003481 x Aromatics x E300;
     47
         In fact, the conditions only guarantee that the  terms in Equation

(D.16) which contain the partial derivatives  (which should be zero at extrema

points)  will be zero.  However, mathematically the two approaches are equivalent.

The regulations were written with the simplest, most straightforward approach to

facilitate implementation of extrapolation.

                                  213

-------
b =  (-0.003626    x Oxygen)   +   (-0.000054    x Sulfur)     +
     (0.043295     x RVP)      +   (-0.013504 x E200)          +
     (-0.062327365 x E300)     +   (0.0282042  x Aromatics)    +
     (-0.002858    x Olefins)  +   (0.000106     x E2002)       +
     (0.000408     x E3002)   +   (-0.000287     x Aromatics x E300;
Phase II Conditions  for  extrapolating exhaust VOC:

     If E200it < 33, then AE200 = E200it - 33
     Otherwise AE200  = 0

     If E300it < 72, then AE300 = E300it - 72
     If E300it > 94 and 79.75 + 0.3846  x ARQt > 94, then
       AE300 = E300it - 94
     Otherwise AE300  = 0

     If AROit < 18, then AARO = AROit - 18
     If AROit > 46, then AARO = ARQt - 46
     Otherwise AARO  = 0

     The extrapolated performance  equations and the associated
conditions for NOx are derived  in  an analogous manner to the
equations and conditions  for VOC.   The results for Phase I and
Phase II are shown below:

Phase I NOx extrapolation equation:
     NOx(t) =
           100% x 0.82 x [(exp(c(et))/l.6437758)  -  1]  +
           100% x 0.18 x [(exp(d(et))/0.8353577)  -  1]  +
           100% x 0.82 x [exp(c(et))/l.6437758] x
           {[-0.000001326 x SULet  +  0.00069205]  x  ASUL +
            [-0.0002381 x AROiet +  0.0083632]  x AARO  +
            [0.00073304 x OLEiet -  0.0027735]  x AOLE} +
           100% x 0.18 x [exp(d(et))/0.8353577] x
           {[-0.000159902 x AROiet  +  0.007097005] x AARO +
            [0.000732 x OLEiet - 0.0027603]  x  AOLE}
c =
[0.0018571
[0.0090744
[0.000846
[-0.002774
  x Oxygen)
  x RVP)
x E300)
 x Olefins)
     [-0.000119 x Aromatics'
d =
[-0.00913
[-0.01397
[-0.00401
[-0.00276
[0.0003665
  x Oxygen)
 x RVP)
 x E300)
x Olefins)
   x Olefins2)
                                +
                                +
                                +
                                +
+
+
+
+
(0.0006921 x Sulfur)      +
(0.0009310  x E200)       +
(0.0083632   x Aromatics)  +
(-0.000000663 x Sulfur2)    +
 (0.0003665    x  Olefins2)
(0.000252
(0.000931
(0.007097
(-0.00007995
              x Sulfur)
              x E200)
           x Aromatics)  +
             x Aromatics2)
                                                               +
                                                               +
Phase I Conditions  for  extrapolating exhaust NOx:

     If SULit < 10, then ASUL = SULt - 10

                                 214

-------
     If SULit > 450, then ASUL = SULit - 450
     Otherwise ASUL =  0

     If AROit < 18, then AARO = AROit - 18
     Otherwise AARO =  0

     If OLEit > 19, then AOLE = OLEit - 19
     Otherwise AOLE =  0
Phase II NOx extrapolation equation:
     NOx(t) =
           100% x 0.738 x [ (exp(c(et))/l.6437758)  -  1]  +
           100% x 0.262 x [(exp(d(et))/0.8353577)  -  1]  +
           100% x 0.738 x [exp(c(et))/l.6437758]  x
           {[-0.000001326 x SULet  +  0.00069205]  x ASUL +
            [-0.0002381 x AROiet  +  0.0083632]  x AARO +
            [0.00073304 x OLEiet  -  0.0027735]  x AOLE}  +
           100% x 0.262 x [exp(d(et))/0.8353577]  x
           {[-0.000159902 x AROiet  +  0.007097005]  x AARO +
            [0.000732 x OLEiet  -  0.0027603]  x AOLE}
c =
[0.0018571
[0.0090744
[0.0008460
[-0.002774
  x Oxygen)
  x RVP)
 x E300)
 x Olefins)
     [-0.000119 x Aromatics'
d =
[-0.00913
[-0.01397
[-0.00401
[-0.00276
[0.0003665
  x Oxygen)
 x RVP)
 x E300)
x Olefins)
   x Olefins2)
+  (0.0006921  x Sulfur)      +
+  (0.0009310  x E200)        +
+  (0.0083632   x Aromatics)  +
+  (-0.000000663 x Sulfur2)   +
 +   (0.0003665   x Olefins2)

+  (0.000252     x Sulfur)     +
+  (0.000931     x E200)       +
+  (0.007097  x Aromatics)     +
+  (-0.00007995 x Aromatics2)  +
Phase II Conditions  for  extrapolating exhaust NOx:

     If SULit < 10, then ASUL = SULt - 10
     If SULit > 450, then ASUL = SULit - 450
     Otherwise ASUL  =  0

     If AROit < 18, then AARO = AROit - 18
     Otherwise AARO  =  0

     If OLEit > 19, then AOLE = OLEit - 19
     Otherwise AOLE  =  0

6.   Response to  Comments

     A number of  comments  were  received concerning the fuel
parameter ranges  that  limit  the  use of the Complex Model which can
be summarized as  follows:
                                 215

-------
     •    The limits on the use of the Complex Model must allow a
          refiner as much flexibility as possible.

     •    The limits on the use of the Complex Model cannot
          preclude legal 1990 fuels or California Phase II
          reformulated gasolines.

Some commenters also suggested that all limits on the use of the
Complex Model should be removed.  As described below, the Agency
determined that such limits were both necessary and expedient.

     The Agency has recognized the need for flexibility in the
reformulated gasoline and anti-dumping programs.  This need for
flexibility was one of the primary reasons for developing an
empirical model that estimates vehicle emissions given the
composition of a given fuel.  Thus the Simple and Complex models
represent a strategy for liberating refiners from the burden of
testing every batch of fuel to determine if it complies with the
emission performance standards.  In this capacity, empirical
models offer an increase in flexibility over testing.

     However, implicit in the development of an empirical model
for vehicle emissions are limits to its applicability.  These
limitations arise from the limitations of the database used to
develop the model.  The most important ramification of the
limitations of the Complex Model database is that the model cannot
be applied reliably to all possible fuel compositions.  Since the
models are based on test data gathered for a limited range of fuel
parameter levels, it would be inappropriate for the Agency to
assume that the Complex Model can be applied without restriction
to all possible fuels.  Further discussion of the limitations of
the Complex Model and the criteria used to establish range limits
can be found in Section D.

     The Agency has made every attempt to set fuel parameter range
limits that do not overly restrict the range of gasolines that can
be evaluated with the Complex Model.  However, the wide variation
in the composition of 1990 gasolines makes it impossible to
designate range limits that simultaneously cover all possible 1990
gasolines and all possible reformulated gasolines while
simultaneously ensuring the reliability of the Complex Model's
emission performance predictions.  Thus the Agency chose to
constrain the range of gasolines covered by the Complex Model in
order to ensure the reliability of the model within those limits.
As discussed previously, EPA specified less restrictive limits on
the use of the model for anti-dumping purposes than for
reformulated gasoline purposes.  The Agency believes that
extension of the valid range of the model beyond these limits
would be inappropriate given the limitations of the underlying
database.  It should be noted that the limits imposed on the range
of gasolines which can be evaluated using the Complex Model do not
impose limits on the range of gasolines which may be produced and

                                216

-------
sold under either the reformulated gasoline or anti-dumping
programs.   The Complex Model may be augmented through vehicle
testing.  In the case of the anti-dumping program,  fuel suppliers
could continue to make a fuel identical to their 1990 baseline
fuel; alternatively,  they could seek approval to produce and sell
fuels that are no worse (based on the emission performance trends
present in the Complex Model) for each parameter.

     Commenters expressed general support for the extrapolation
methodology proposed by the Agency.  However, the increased
complexity in the Complex Model equations introduced through
extrapolation led some commenters to suggest that the regression
curve approach should be used in all but the most extreme cases
(see Section D.3 above for details on the various extrapolation
approaches).   The Agency has determined that the final
extrapolations presented in this section provide the most
appropriate approach to contending with the limitations of the
Complex Model database for the following reasons:

     •    Applying linear extrapolation to all second order terms
          avoids arbitrary determination on the part of the Agency
          as to which terms exhibit unacceptably "extreme"
          behavior.

     •    The limitations of the database, as defined by the data
          core, are used to determine the points at which
          extrapolation should begin.  Thus the definition of the
          data core provides an appropriate basis for defining
          points at which extrapolation should begin.

Some commenters also provided suggestions for the manner in which
to extrapolate specific model equations.  For the most part, these
suggestions applied to the various Complex Model options which
were not adopted for the final Complex Model.  The Agency has
taken the specific extrapolation suggestions that applied to the
final Complex Model equations into consideration.


E.    Use of the Complex Model Under Non-Summer Conditions

     While the VOC performance standard for reformulated gasolines
applies only in the summer, the toxics and NOx requirements apply
year-round.   Similarly, the exhaust toxics and NOx requirements of
the anti-dumping program apply year-round.  EPA therefore
recognizes that the complex model must be extended to cover
typical winter fuels in order to estimate the exhaust toxics and
NOx emission performance of gasolines during the winter months.
Modeling winter emissions performance presents a number of unique
difficulties.  These difficulties include:

     •    The data sources described in Section A.I provide data
          on emissions performances only under summer conditions.

                                217

-------
     •    The wide range of temperatures encountered during the
          winter season (September 16th to March 30th)  in the
          various geographical areas (north vs. south)  within the
          United States would make emissions testing and modeling
          under winter conditions more difficult.  The range of
          temperatures during the summer season tend to be
          considerably more uniform, particularly within each VOC
          Control Region.

     •    The RVP levels of fuels included in the current complex
          model database range from 7 to 10 psi, while winter
          fuels tend to have RVP levels in the range of 11 to 13
          psi .

For these reasons, the complex model described in Section A cannot
be used directly to evaluate fuels with typical winter RVP levels.
The method chosen by EPA for extending the complex model for
determining performance during wintertime scenarios is described
below for both exhaust as well non-exhaust emissions.

1.   Non-Summer Exhaust Model

a.   Winter Engine Operation

     The effect of fuel modifications on exhaust VOC and NOx
emissions are thought to be minimally affected by the cooler
ambient wintertime temperatures.  Auto/Oil data  [A-92-12 Docket
submittal #II-G-1, December 1990] has shown that most fuel
parameters affect emissions in two ways as the fuel is burned in
the engine:

     •    Fuel properties affect the quantity of exhaust emissions
          generated in the engine itself (i.e., an "engine-out"
          effect) .

     •    Fuel characteristics affect the efficiency of the
          catalyst in reducing the level of emissions leaving the
          vehicle (i.e., a "catalyst" effect).

Most fuel parameters affect both engine-out emissions and catalyst
efficiency.   Fuel sulfur content is the one fuel parameter which
is believed to have only a catalyst effect.  It is expected that
fuels will encounter almost identical conditions in the combustion
chamber  (resulting in almost identical engine-out effects) and the
catalyst (resulting in almost identical catalyst effects) during
most of the driving cycle regardless of the ambient temperature.
The only differences between summer and winter engine and catalyst
operating conditions are:

     •    Slightly cooler engine temperatures due to increased
          heat transfer rates from the engine to the cooler winter
          air.

                                218

-------
     •    Slightly longer warm-up times for engine and catalyst
          due to the thermal inertia of the colder engine block,
          coolant,  and catalyst as well as the increased heat
          transfer rates to the ambient environment during warm-
          up.

     •    Slightly longer open-loop engine operation resulting
          from the longer warm-up time.

     •    Lower vapor generation rates and reduced canister
          loading due to lower fuel temperatures and lower ambient
          air temperatures.

Except for the last point,  these differences are not expected to
heavily influence the emission effects of most fuel parameters in
winter when compared summer conditions.  Hence EPA has concluded
that most fuel modifications should have similar exhaust emission
effects in both summer and winter, despite potentially large
differences in ambient temperatures during the two seasons.

     This conclusion is supported by data available from the
Auto/Oil studies which has shown that the effect of most fuel
modifications on exhaust emissions is similar regardless of
temperature.  Bagl data taken on the FTP reflects measurements
taken during the cold phase of engine start-up while bag3 data
reflects measurements taken after the engine heats up to running
conditions.  Thus,  a comparison of the effects of fuel
modifications on exhaust emission data from bagl and bag3 should
serve as an indicator of how temperature effects emissions.  Data
from Auto/Oil [A-92-12 Docket submittal #II-G-1, December 1990]
has shown many fuel modifications to result in similar exhaust
emission effects in both bagl and bag3.   While the Auto/Oil data
confirms EPA's expectation that many fuel modifications would have
similar effects under summer and winter conditions, the amount of
data provided by the Auto/Oil study is too limited to form the
basis of a separate winter model.

     EPA does not believe that this conclusion can be applied to
RVP, however.  As indicated previously, vapor generation rates
under summer conditions are substantially higher than under winter
conditions for fuels with identical RVP levels.  EPA believes that
RVP's effects on emissions are due in part to RVP's impact on
actual vapor generation in the engine.  However, the Agency
believes that RVP's primary effect on emissions arise due to the
loading and subsequent purging to the engine of the evaporative
emissions control canister.  Both phenomena are highly
temperature-sensitive; fuels with identical RVP levels generate
far more vapor and produce far greater canister loading under
summer conditions than under winter conditions.  Vapor generation
and canister loading are a function of a fuel's true vapor
pressure, which increases as temperature increases in accordance
with basic principles of gas-phase chemistry.  Hence EPA is not

                                219

-------
confident that RVP effects estimated from test data gathered under
summer conditions can be used to estimate RVP effects under winter
conditions.
b.
Winter Baselines and Fuels
     The winter baseline emissions of VOC and NOx are derived
directly from the appropriate MOBILE models as outlined in Section
III.B.  Winter baseline toxic emissions are derived from the
Complex Model database and the Complex Model itself.   All winter
baseline emissions are determined using the following fuel
parameters,  as discussed further in Section III.B:

                           Table IV-49
           Baseline Fuel Parameters Used in Determining
                     Winter Baseline Emissions
Fuel Parameter
Oxygen
(weiqht percent)
Sulfur
(ppmW)
RVP
(psi)
E200
(percent)
E300
(percent)
Aromatics
(volume percent)
Olef ins
(volume percent)
Benzene
(volume percent)
Winter Baseline Value
0.00
338.0
11.5
50.0
83.0
26 .4
11.9
1.64
     Winter fuel characteristics are comparable to summer fuel
characteristics except for Reid Vapor Pressure (RVP).   The higher
winter RVP levels are needed to maintain similar actual vapor
generation rates under winter despite the generally lower fuel
temperatures under such conditions (particularly during engine
warm-up, when emissions are highest).   As discussed previously,
the RVP levels of fuels included in the exhaust complex model
database ranged from 6.5 to 10 psi, while winter fuels tend to
have RVP levels in the 10-14 psi range.  RVP's impact  on canister
loading and purge rates are thought to be the primary  causes of
its effect on exhaust emissions.  Since data on the emission
effects of fuels with winter RVP levels under winter conditions
                                220

-------
are extremely limited, EPA has not been able to model the effects
of winter RVP levels on exhaust emissions.
c.   EPA Solution

     Based on the points discussed in the previous section, EPA
proposed in the February 1993 NPRM that for purposes of evaluating
exhaust emission effects using the complex model, the RVP of
winter fuels be set at the summer statutory baseline RVP value.
EPA did not receive comments opposing this approach, which the
Agency believes avoids the use of the model to make unfounded
predictions.  EPA has incorporated this approach in the final
reformulated gasoline and anti-dumping requirements, which has the
effect of embedding into the Complex Model the assumption that the
RVP level of winter gasolines has no effect on NOx or exhaust
toxic emissions.  Thus,  to evaluate wintertime fuels using the
complex model, an RVP value equal to that of summer baseline
gasoline (8.7 psi)  must be used instead of the fuel's actual RVP.
This procedure effectively removes the contribution of RVP to
winter exhaust emissions.  By removing any emissions-related
incentives to alter winter RVP levels from 1990 levels, this
approach also reduces the likelihood that in-use winter RVP
levels, or the associated emissions, will change from their 1990
levels.  To the extent that such emissions do not change, the
approach outlined above becomes even more appropriate.

2.   Non-Summer Non-Exhaust Model

a.   NPRM Approach and Comments

     In all its prior proposals, the Agency had proposed that
winter nonexhaust emissions, including winter nonexhaust benzene
emissions,  be set equal to zero.  EPA received a number of
comments requesting that both baseline emissions and the
nonexhaust toxics model include winter nonexhaust benzene
emissions.

     These requests were based on the fact that the year-round
benzene limits would result in reduced nonexhaust benzene
emissions in the winter months, if such emissions existed.  EPA
has evaluated this claim, taking into account winter temperature
ranges and the effects of different inspections and maintenance
programs on such emissions.  EPA believes that this claim is
valid, since winter non exhaust emissions (including nonexhaust
benzene emissions)  are unlikely to be zero under all winter
temperature ranges.

     In the past, the absence of sufficient data on nonexhaust
emissions under winter temperature conditions has prevented EPA
from developing accurate models of winter nonexhaust emissions.
The commenters provided a limited amount of data on winter

                                221

-------
nonexhaust emissions to support their claim.  However,  the data
submitted was found to contain the following limitations:

     •    The data submitted in support of this claim were based
          on measurements of nonexhaust emissions from vehicles
          with very low nonexhaust emissions that are not
          representative of actual in-use vehicles.

     •    The chemical composition of the measured nonexhaust
          emissions were characteristic of resting losses rather
          than diurnal, hot soak or running loss emissions.
          Resting losses are not included in EPA's baseline
          emission estimates,  so the Agency does not consider it
          fair to include resting losses in its nonexhaust
          emission models.  Speciation of the nonexhaust emissions
          also revealed the presence of compounds not found in
          gasoline, suggesting that non-fuel sources of emissions
          were present.

     •    No data were submitted on nonexhaust benzene emissions
          from fail vehicles under winter conditions.  Since
          nonexhaust benzene emissions from such vehicles will
          comprise a large portion of winter nonexhaust benzene
          emissions, EPA is concerned that a model based only on
          pass vehicles would not provide accurate estimates of
          winter nonexhaust benzene emissions.
b.   EPA Decision

     For the reasons discussed in the previous section,  the data
submitted by the commenters on winter nonexhaust emissions cannot
be used to evaluate wintertime baseline emissions.  Thus,  EPA is
promulgating models which assume that winter nonexhaust emissions,
including winter nonexhaust benzene emissions, be considered zero.
However, given the merit of the claims made by the commenters,  EPA
will consider including a model of winter benzene nonexhaust
emissions in the complex model in the future when sufficient
reliable data become available.
F.    Behavior of Complete Complex Model

1.    Fungibility

     As the gasoline distribution system currently operates,  all
gasoline is fungible.  Fungibility means that a given batch of
gasoline is exchangeable and/or mixable with any other batch of
gasoline at any point between the refinery gate and a vehicle's
                                222

-------
fuel tank48.  Fungibility  simplifies  gasoline distribution because
it permits gasoline to be stored, transported,  and  sold  without
the constraints and costs associated with  segregating  every batch
of gasoline from every other batch.  Thus  gasoline  is  always
gasoline regardless of whether it derives  from  a mixture of other
gasolines.

     Reformulated gasoline  (RFC) will be distributed through the
same network of pipelines,  trucks, and terminals that  is currently
used for conventional gasoline.  However,  because RFC  is produced
to very different specifications than conventional  gasoline,  the
production of RFC potentially will have significant impacts on the
fungibility of the gasoline distribution system.  These  impacts
fall into two categories:

     1)   The difference  in production specifications  for
          conventional gasoline and RFC require that all
          conventional gasoline must be segregated  from  all
          RFC at all times.

     2)   The compliance  of a given batch  of RFC with  the
          emission standards is determined through  the
          Complex Model which, in general,  is not linear.
          The result of this nonlinearity  is that the
          emissions performance of a mixture of two complying
          RFGs cannot be  easily inferred from the
          performances of the individual fuels.  It may  be
          possible, therefore, that the mixture of  two
          complying RFGs  will yield a fuel that is  not a
          complying RFC,  according to performance estimates
          provided by the Complex Model.

     The first impact on  fungibility, that of segregation of all
RFG from all conventional gasoline, cannot be compromised.   The
compliance of RFG is determined through evaluation  of  its fuel
parameters with the Complex Model.  If a batch  of RFG  were to be
mixed with a batch of conventional gasoline,  the value of its fuel
parameters could change dramatically.  The "mixed"  RFG would then
require re-evaluation with  the Complex Model, and would  be
unlikely to comply with the requirements of the reformulated
gasoline program.  To assure that conventional  and  reformulated
gasoline is not mixed, conventional gasoline is required to
contain a chemical marker.  This marker is not  permitted to be
present in gasoline sold  as reformulated gasoline.

     The second impact on fungibility requires  both elaboration
and analysis.  The practical and financial requirement that RFG  be
     48
        Some segregation of  conventional gasolines already occurs based on fuel


RVP, octane rating, and/or oxygenate type and content specifications.


                                223

-------
fungible throughout its distribution system requires that RFC
batches be allowed to mix with one another in any proportion and
at any point between the refinery gate and a vehicle's fuel tank
(in practice, the different standards and emission models
applicable in VOC Control Regions 1 and 2 require RFC for each
region to be segregated).  Thus the requirement for a fungible
distribution system translates into a requirement that a mixture
of two complying RFGs must be a complying RFC.  Since RFC
compliance is determined through the Complex Model which is not
linear, it cannot be assumed that mixtures of complying
reformulated gasolines are themselves complying reformulated
gasolines.  To determine the degree to which certified
reformulated gasolines can be mixed and still remain in
compliance, an analysis of the behavior of the Complex Model was
conducted.

     The nonlinearity in the Complex Model is due to the existence
of second-order terms and the exponential form of the regression
equations.  This nonlinearity means that emissions from a mixture
of two fuels will differ from the weighted average of the two
fuels' individual emissions.  However, this mixing may be either
beneficial or detrimental to an RFG mixture's compliance status.
In other words, the Complex Model may predict any of the following
for a mixture of two complying RFGs:

     1)   The emissions performance of the mixture equals the
          weighted average of the two original fuels'
          emissions performances.

     2)   The emissions performance of the mixture is better
          than the weighted average of the two original
          fuels' emissions performances.

     3)   The emissions performance of the mixture is worse
          than the weighted average of the two original
          fuels' emissions performances, but the mixture
          remains a complying RFG.

     4)   The emissions performance of the mixture is worse
          than the weighted average of the two original
          fuels' emissions performances, and the mixture
          becomes a non-complying RFG.

Fungibility problems arise only for case (4).

     The fungibility analysis involved evaluating the performance
of thousands of randomly generated fuels with the Complex Model.
Emission estimates for total VOC, NOx, and total toxics were
calculated to determine the frequency of occasions in which the
mixture of two complying RFGs produced a non-complying RFG.  The
analysis also provided information on the frequency of occasions
in which a mixture of two complying RFGs performed more poorly

                                224

-------
than could be inferred from the two original fuels, but
nevertheless remained a certifiable RFC  (case 3).  The steps  in
the analysis can be summarized as follows:

     1.   Create 24,000 fuels distributed randomly throughout the
          fuels space defined by the valid range of the
          extrapolated Complex Model.

     2.   Retain all "certifiable" and realistic fuels based  on
          VOC performances49 of -15.0 to  -17.0 percent, NOx
          performances of 0.0 to -2.0 percent,  and toxics
          performances of -15.0 to -30.0 percent.  Fuels with
          emission performance levels beyond these ranges  are
          either not permissible under a per-gallon compliance
          scenario or are unrealistically low-emitting.  Including
          such fuels makes fungibility problems  (in which  the
          mixture does not comply with the per-gallon
          requirements) even less likely.

     3.   Generate 50:50 mixtures of pairs of fuels from among
          those retained from step 2 by averaging the fuel
          parameter values from the two contributing fuels.

     4.   Determine the Complex Model performance of the mixtures
          generated in step 3 for VOC, NOx, and toxics.

     5.   Calculate the "Fungibility Effect" by subtracting the
          average performance of the two constituent fuels from
          the performance of the mixture  (the average performance
          of the two constituent fuels is equivalent to the
          weighted average of the performances  of the two  fuels
          mixed in a 50:50 ratio):

          Fungibility Effect = M -  [(0.5 x A) +  (0.5 x B)]

          M =  Performance of the mixture  (% change from baseline)
          A =  Performance of the first qualifying RFG  (%  change
               from baseline)
          B =  Performance of the second qualifying RFG  (% change
               from baseline)

          Fungibility effects were determined separately for  each
          pollutant.  Note that negative fungibility effects  are
          beneficial, i.e. negative fungibility effects indicate
          that emissions from the mixture are lower than would
          have been expected based on the performance of the  two
          original fuels.
     49
        Performance is here defined as the percent change in emissions of a


target fuel  as compared to a baseline fuel


                                225

-------
     6.   Examine the average fungibility effect to determine the
          impact of fuel mixing on in-use compliance and the
          potential application of ratchets to the applicable
          emission standards.

     7.   Examine fuels for which the fungibility effect was
          positive (i.e.,  detrimental)  to identify situations that
          might result in the issuance of a Notice of Violation.

     8.   Determine cases in which a detrimental fungibility
          effect caused a mixture of two complying fuels to be
          non-complying.

     The scenario under which the analysis was performed was
designed to amplify any fungibility problems that might exist.
For this reason, a per-gallon scenario with tight controls on the
initial performance estimates was assumed, along with a 50:50 fuel
mixing ratio.  Averaging scenarios are less binding since NOVs can
be issued only for violations of the per-gallon minimum
performance requirements,  and because the wide range of fuels
likely to exist under averaging are unlikely to consist only of
fuels with unfavorable fungibility effects.  The 50:50 mixing
ratio tends to create the largest fungibility effects, since any
nonlinear effects are most noticeable at this ratio.  Fungibility
effects and problems are independent of the program phase, area
class, and season assumed for the analysis, since the percentage
change estimates for the original fuels and mixtures are largely
independent of these conditions.  Oxygenate type affects emissions
only in the case of the aldehydes and nonexhaust benzene, since
such effects are linear, the analysis is independent of oxygenate
type.  The scenario analyzed below was based on Phase I summer
requirements in Class C areas and used MTBE as the oxygenate.  The
results of the analysis are shown in Figures IV-2 through IV-7,
and are summarized in Table IV-50.
                                226

-------
Figures IV-2 and IV-3:   Results of VOC Fungibility Analysis
                                227

-------
Figures IV-4 and IV-5:   Results of NOx Fungibility Analysis
                                228

-------
Figures IV-6 and IV-7:   Results of Toxics Fungibility Analysis
                                229

-------
Table IV- 50
Summary of Results of Fung

Performance of
contributing fuels
Minimum
Maximum
Average
Performance of mixtures
Minimum
Maximum
Average
Fungibility effects
Minimum
Maximum
Average
Total VOC

-16.97
-15.00
-16.01

-17.77
-15.20
-16.42

- 1.81
1.01
- 0.42
ibility Analysis
NOx

-1.98
0.00
-0.93

-3.20
0.43
-0.99

-1.68
0.86
-0.04
Total toxics

-29.99
-15.49
-25.06

-29.70
-20.06
-25.40

- 1.19
0.02
- 0.35
     A number of conclusions can be drawn from the values in Table
IV-50.  First, the minimum and maximum performances of the
contributing fuels fall within the performance ranges given in
step 2, since only complying and realistic fuels were retained
from the 24,000 randomly generated fuels.  Second, the average
performance of the mixtures is better than the average performance
of the contibuting fuels for total VOC,  NOx,  and total toxics,
which indicates that, on average, mixing fuels tends to result in
beneficial fungibility effects.  Other conclusions are discussed
separately for VOC, NOx, and toxics:


Total VOC

     The average VOC fungibility effect is beneficial, although a
small percentage of such effects are detrimental.  However,  in no
case does a fungibility effect result in a mixture of two fuels
having emission reductions of less than 15.0%.  Thus this analysis
shows that although some mixtures of pairs of fuels will have
higher emissions than would be implied by the performances of the
two constituent fuels,  these higher emissions will never result in
a mixture of two complying fuels being in non-compliance with the
standards for RFG.  This conclusion also applies to compliance on
average.  When complying on average, the performance of individual
fuel blends will cover a larger range than when complying on a
per-gallon basis.  EPA has found that the distribution of VOC
fungibility effects becomes more beneficial as the range of
performances increases.  In addition, EPA found that detrimental
fungibility effects are not large enough to result in violations
                                230

-------
of the per-gallon minimum performance standards under averaging.
Finally, the compliance margin imposed under averaging assures
that fuels complying on average will result in average VOC
performances that exceed the per-gallon standards applicable in
each Phase.
NOx

     The average NOx fungibility effect is beneficial,  although a
significant percentage of mixtures resulted in detrimental
fungibility effects.  EPA found that the vast majority of mixtures
with detrimental effects comply with the program's NOx
requirements.  Of the 124 fuel pairs that survived the screening
process of step 2,  only two pairs resulted in detrimental
fungibility effects that also resulted in NOx noncompliance for
the mixture.  These two fuel pairs,  their mixtures, and the
Complex Model performances for each fuel are shown in Table IV-51.
Table IV-51
Fuels Resulting in Non-com


Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Benzene
Performance :
Total VOC
NOx
Total Toxics
Fuel
#1
3.19
172
8.17
38.4
91.5
44.2
17.4
0.371

-15.26
- 0.22
-27.60
Fuel
#2
2 .21
399
8.04
35.2
93.4
24.5
9.15
0.843

-15.49
- 0.99
-27.37
#1 +
#2
2 .70
285
8.11
36.8
92 .4
34.4
13.3
0.607

-15.37
0.22
-27.79
plying Mixtures
Fuel
#3
3.29
420
8.14
45.9
90.4
22 .2
9.66
0.968

-15.50
- 0.06
-29.55
Fuel
#4
2 .48
291
8.08
57.0
79.3
35.6
7.63
0.739

-15.07
- 0.46
-26.85
#3 + #4

2.88
356
8.11
51.5
84.9
28.9
8.65
0.853

-16.24
0.44
-28.38
     Such results show that the mixture of two complying RFGs can
result in a non-complying gasoline for NOx.  However,  the
following points indicate that such concerns are not warranted:

•    The scenario under which the analysis was carried out was
     designed to amplify fungibility effects and any associated
     fungibility problems.  Under actual in-use conditions, these
     effects would be substantially smaller or nonexistent since
     50:50 ratios of the two fuels are unlikely and since fuels
     other than the fuels presented above are likely to be mixed.
     As more fuels are mixed, fungibility effects will tend to
                                231

-------
     reflect the average fungibility effects presented in Figures
     IV-2 through IV-7.

•    None of the four constituent fuels in Table IV-51 are
     representative of in-use reformulated gasolines:

     •    Low sulfur does not typically accompany high olefins,
          and vice-versa, as in Fuels #1 and #3.

     •    Low E200 does not typically accompany high E300 as in
          Fuels #1, #2, and #3.

     •    Ultra-high octane fuels, such as Fuel #1,  will most
          likely not play a significant role in the reformulated
          gasoline program.  In addition,  such fuels are unlikely
          to be mixed with low-octane fuels such as fuel #2 due to
          the price premium commanded by high-octane fuels and the
          cost of the additional oxygenate.

•    Very few, if any, refiners will produce RFGs whose
     performances fall as close to the standards as those in Table
     IV-51 do.  To assure compliance with the per-gallon standards
     given the uncertainties associated with measuring fuel
     parameter levels, fuel suppliers are likely to maintain
     compliance margins.

•    The oxygenate contents are not at the levels commonly
     associated with MTBE or ethanol fuels.  In fact, fuels #1 and
     #3 can be produced only with ethanol as the oxygenate.

Thus under in-use conditions, EPA does not anticipate any NOx
fungibility problems to arise under per-gallon compliance with the
"no NOx increase" provisions of the program.  As discussed
previously,  fungibility effects under averaging are likely to be
more beneficial than under per-gallon compliance since the range
of performances of fuels produced under averaging will be wider
than the range of performances of fuels produced under per-gallon
compliance.   In addition, EPA found that detrimental fungibility
effects are not large enough to result in violations of the per-
gallon minimum performance standards under averaging.  Finally,
the compliance margin imposed under averaging assures that fuels
complying on average will result in average NOx performances that
exceed the per-gallon standards applicable in each Phase.


Total Toxics

     All exhaust toxics, except for POM, have strictly linear
relationships with all fuel parameters, and therefore would not be
expected to demonstrate either beneficial or detrimental
                                232

-------
fungibility effects.50  The non-exhaust benzene  models,  however,
include a negatively squared relationship with  RVP,  and therefore
would be expected  to exhibit beneficial fungibility  effects at all
times.  These  facts  are confirmed in the results  for total  toxics,
as the fungibility effects are almost always beneficial (though
POM could contribute a small detrimental fungibility effect).
Thus it is apparent  that all 50:50 mixtures of  complying and
realistic fuel pairs also will comply and in fact will  almost
always have lower  emissions of toxics than the  weighted average of
the contributing fuels'  toxics emissions.

2.    Predictability  and Validation of Exhaust Complex Models

     As discussed  in the February 1993 NPRM and in previous EPA
submittals to  the  docket,  there are two ways by which the accuracy
and/or precision of  models can be investigated.   One way is to
inspect the effect of changing one fuel parameter at a  time on
exhaust emissions  and compare the effect to directional effects
demonstrated by specific studies in the database.  Another  way is
to compare model predictions on vehicles and fuels from specific
test programs  to measured emission effects of fuel modifications.
Both of these  methods will be discussed in this section.
a.   Parametric  Performance of Complex Models

     Table IV-52  summarizes the effects of single parameter fuel
modifications on  exhaust emissions as predicted by  the  final
(simplified) Phase  I  Complex Models  (see Section A.9) for the 1990
in-use fleet.  With the exception of the poorly-understood effects
of E200 fuel modifications on exhaust NOx and VOC emissions,  the
emission effects  predicted by the Complex Models for  exhaust VOC
and NOx are directionally consistent with the effects predicted by
the Auto/Oil and  ATL  studies [See A-92-12 Docket submittals II-G-1
through II-G-5, December 1990;  II-B-12, February 1993 and II-B-14,
February 1993].   Since  emissions data from the Auto/Oil and ATL
studies comprise  over 50% of the emissions database,  the Complex
Models appear to  demonstrate behavior that is directionally
consistent with  the data in the Complex Model database.   As
discussed in Sections III and IV.A, there are far fewer data on
        Since the  exhaust models are in the form of exponentials of

polynomials,  there is in fact some curvature to "strictly linear" relationships

that can contribute  to fungibility effects.  However, the nature of an

exponential function is such that only beneficial effects would be exhibited, and

the relatively small order of magnitude of the independent variables (fuel

parameters) is such  that the exhibited effects would be very small.

                                 233

-------
toxic emissions than on VOC and NOx emissions.  Hence the models'
predictions for toxic emissions behavior are expected to be less
accurate than the models' predictions of VOC and NOx emission
behavior.    However, the effects of fuel modifications on exhaust
toxic emissions shown in Table IV-52 below are also, for the most
part, directionally consistent with the studies that measured and
quantified the effects of fuel parameter changes on toxic exhaust
emissions  [See A-92-12 Docket submittals II-G-1 through II-G-5,
December 1990; II-B-12,  February 1993 and II-B-14, February 1993].

                            Table IV-52
     Phase I,  Summer, In-Use Exhaust Complex Model Predictions
    (all  numbers are  percent  change  in  emissions  from CAAB  fuel)
Fuel
Parameter
Changed
Oxygen
0->2 wt%
Sulfur
339->139 ppm
RVP
8.7->7.7 psi
E200
41->51 %
E300
83->87.3 %
Aromatics
32->27 vol%
Olef ins
9.2->4.2 vol%
Benzene
1.53->0.95
vol%
VOC
&
POM
-0.72
-4.63
-3.52
-4.11
-2 .64
-1.94
+ 1.44
	
NOx
-0.02
-6.81
-0.49
+ 0.94
-0.01
-0.76
-1.06
	
Benzene
-8.39
-9.19
	
-1.73
+ 2.58
-9.23
	
-12 .1
Butadiene
-5.49
-1.59
	
-7.02
-4.85
+ 2 .02
-16.3
	
Acet-
aldehyde
-6.07*
-5.12
-2.03
	
-5.10
+ 2 .81
	
	
Form-
aldehyde
+9.68*
	
	
	
-4.30
+ 3.65
+ 8.15
	
  Calculated assuming  oxygen in the  form of  MTBE
b.   Validation of Complex Models

     Because of the wide range of fuel parameters tested and the
large number of vehicles tested in both the normal and higher
emitter categories, EPA's ATL-I and ATL-II programs were chosen
for comparison to model predictions of VOC and NOx emissions.
Selected fuels with a wide range of different fuel properties,
emissions effects, and oxygenate types) from the various Auto/Oil
programs were also chosen for comparison to the exhaust Complex
Models developed above.  The fuels from the different programs and
their properties are summarized below in Tables IV-53, IV-54 and
IV-55.
                                234

-------
                   Table IV-53
ATL-I  Fuels  Used  for  Validation  of  Complex  Model
Fuel
C
D
E
F
G
H
Oxygen
(wt %)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
3.5
(ETOH)
Sulfur
(ppmw)
371
364
355
112
133
353
RVP
(psi)
8.3
7.6
8.5
8.3
7.7
8.0
E200
(%)
57.1
54.1
62 .2
54.8
61.8
57.8
E300
(%)
85.0
83.8
92.5
83.9
92.3
84 .7
Arom
(v %)
25.4
25.8
23.6
24 .2
23.8
27.3
Olef
(v %)
6.8
6.4
5.9
6.4
5.6
7.0
Benz
(v %)
1.0
1.0
0.9
0.9
1.0
1.0
                   Table IV-54
ATL-II Fuels Used for Validation of Complex Model
Fuel
K
L
M
N
P
R
S
T
U
V
Oxygen
(wt %)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
2.0
(MTBE)
Sulfur
(ppmw)
311
307
345
337
288
320
327
59
66
31
RVP
(psi)
7.5
7.7
7.3
7.7
7.9
7.9
7.5
7.6
6.4
6.2
E200
(%)
48.9
52.3
42 .6
52.4
53.5
53.6
52.4
52.9
52.6
50.5
E300
(%)
80.0
81.8
81.7
76 .2
83.7
81.1
81.6
81.4
81.3
90.8
Arom
(v %)
28.9
26 .4
25.7
23.1
26.8
25.0
23.3
22 .8
23.5
21.7
Olef
(v %)
7.0
7.0
7.3
7.8
9.8
7.4
1.1
1.3
1.4
2.8
Benz
(v %)
0.8
0.8
0.9
0.6
0.8
0.6
0.5
0.5
0.5
0.8
                       235

-------
                            Table IV-55
        Auto/Oil Fuels Used for Validation of Complex Model
Fuel
H
(AMOT)
0
(AMOT)
R3A
(HC/A)
MM
(RVP)
N2
(RVP)
NN
(RVP)
T
(RVP)
Oxygen
(wt %)
2 .7
(MTBE)
2 .7
(MTBE)
2.0
(MTBE)
2 .7
(MTBE)
2 .7
(MTBE)
2 .7
(ETBE)
3.5
(ETOH)
Sulfur
(ppmw)
313
288
90
345
313
313
246
RVP
(psi)
8.5
8.6
8.48
8.0
8.8
8.0
9.3
E200
(%)
65.3
47. 9
46 .2
67.4
65.7
50.7
57.4
E300
(%)
92.0
91.9
87.1
91.3
91.3
92.6
93.1
Arom
(v %)
20.2
46.7
28.2
22 .2
22 .5
21.2
18.1
Olef
(v %)
20.2
19.3
14.5
5.4
4 .7
3.8
3.6
Benz
(v %)
1.52
1.38
0.62
1.41
1.31
1.50
1.35
     The exhaust emissions data (after converting total
hydrocarbons to non-methane hydrocarbons in order to facilitate
direct comparisons to percent change predictions from the Complex
Models)  from these programs were placed in appropriate technology
groups and emitter classes and re-weighted according to the
revised weighting factors listed in Section III.B.4 for summer
Phase II VOC and NOx emissions.  Shown in Table IV-56 is a direct
comparison of data to predicted exhaust emissions from the final
Complex Models for VOC and NOx.  The percent change in emissions
shown in Table IV-56 for the data as well as predictions by the
model (which are in bold type) should be directly comparable since
all weightings and classifications used were identical.
                                236

-------
                            Table IV-56
         A Comparison of Data to Complex Model Predictions
Study /Fuel
ATL-I/C
ATL-I/D
ATL-I/E
ATL-I/F
ATL-I/G
ATL-I/H
ATL-II/K
ATL-II/L
ATL-II/M
ATL-II/N
ATL-II/P
ATL-II/R
ATL-II/S
ATL-II/T
ATL-II/U
ATL-II/V
AO/H
AO/0
AO/R3A
AO/MM
AO/NN
AO/N2
AO/T
percent change in VOC
emissions from base fuel
NORMAL
EMITTERS
-10.4/-8.44
-13.4/-8.77
-14.2/-11.6
-14.5/-18.9
-17.8/-22.8
-9.4/-10.0
-4.6/-3.S
-10.2/-7.3
-5.S/-2.9
-1.2/-0.97
-12.2/-10.3
-10.4/-6.4
-4.9/-6.S
-12.4/-18.1
-12.8/-20.2
-22.5/-25.S
-17.8/-18.4
-15.0/-15.6
+1.39/-20.9
-10.1/-14.7
-8.10/-13.9
-7.10/-13.1
-7.30/-14.9
HIGHER
EMITTERS
-13.3/-9.4
-10.8/-11.3
-7.90/-8.7
-14.7/-8.0
-11.4/-10.5
-20.3/-10.4
-1.7/-7.6
-S.3/-9.4
-3.1/-8.5
-3.4/-S.9
-4.3/-9.S
-2.6/-9.S
+1.1/-10.1
-3.1/-8.7
-4.3/-13.0
-2.7/-13.9
*
	
	
	
	
	
	
percent change in NOx
emissions from base fuel
NORMAL
EMITTERS
-2.5/+1.5
-2.9/+0.35
-3.7/+1.8
-4.8/-S.4
-8.5/-6.S
+5.7/+1.6
-0.36/-0.28
-1.3/-0.19
-4.7/-0.54
-4.1/-0.45
-3.9/+0.72
+1.2/+0.20
-5.8/-0.95
-10.6/-11.7
-11.2/-12.1
-10.2/-14.1
+10.6/+7.8
+3.31/+8.2
+3.S9/-6.3
+9.02/+0.31
+13.3/+0.09
+6.39/-2.3
+12.0/-3.6
HIGHER
EMITTERS
-2.4/-1.4
+1.5/-0.4
-9.9/-S.2
-11.5/-7.8
-10.2/-9.3
+3.5/-2.0
+5.9/+0.2
+2.6/-1.2
+6.7/-0.7
-1.6/+0.9
+1.1/-1.6
+0.9/-1.1
-1.6/-1.7
-2.8/-S.3
+5.6/-S.3
+4.8/-11.1
*
	
	
	
	
	
	
  Higher  emitter  were  not  tested  in  the Auto/Oil  Programs

     For the most part,  it is evident from Table IV-56 that the
discrepancies between actual test results and model predictions
                                237

-------
are small and would be expected to fall within the 90% confidence
interval calculated in the ATL test programs  [see A-92-12 Docket
submittals #II-B-12, February 1993 and #II-B-14,  February 1993].
In some cases, the predictions are poor due to contradicting
emissions for similar fuels within the data in the Auto/Oil and
ATL studies.

3.    Response to Comments

     In response to previous EPA submittals to the docket, one
commenter expressed the need for the exhaust benzene model to be
developed in a different manner.  The commenter suggested that the
effect of fuel benzene content and fuel aromatics content were
overestimated.  However, the agency has analyzed many exhaust
benzene emission models  [see EPA memorandum by Rao and Sklar
"Alternative Exhaust Benzene Models," December 2, 1993], including
the model suggested by this commenter, and concluded that the
exhaust benzene model presented in Section A is the simplest in
form and as accurate as any other model for exhaust benzene
emissions.  EPA, in all of the models that it has developed for
predicting exhaust benzene emissions, has found that fuel benzene
content and aromatics content have a large effect on benzene
emissions, particularly when their effects on emissions from
vehicles with emission levels representative of in-use emission
levels.

     Nearly all commenters expressed the need for a fungible fuel
distribution system.  The Agency agreed with commenters that
fungibility is a critical element of the fuel distribution system.
EPA evaluated fungibility throughout the development of the
Complex Model.  The fungibility analysis methodology described in
Section F.I was originally presented at the June 2, 1993 Complex
Model workshop in Romulus, Michigan in response to the need for a
fungible model.  It has since been affirmed by the participants of
that workshop as an appropriate and accurate analysis tool.

     In response to the July 15, 1993 docket submission of various
Complex Model options, commenters concluded that the Agency was
fulfilling its mission to develop a Complex Model that ensured
fungibility.  The Agency has continued this mission in the
development of the final Complex Model, as the results of the
fungibility analysis in Section F.I show.  Thus the EPA's analysis
shows that, for real in-use fuels, the final Complex Model
presents no fungibility concerns.
                                238

-------
G.   Vehicle Testing

     As discussed in Section V of the preamble,  the purpose of
vehicle testing is to supplement the Complex Model, not to
supplant it.  Vehicle testing is subject to a number of sources of
error and bias.  The model is less vulnerable to bias and error
since it is based on the results from a number of independent
testing programs.  Nonetheless,  the results of new vehicle testing
programs can provide information beyond that available from the
test programs represented in the Complex Model database.   New
vehicle testing programs are particularly valuable when they
examine the effects of fuel parameters at previously-untested
levels or the effects of fuel parameters that have not yet been
tested.

     In order to provide useful information beyond that
incorporated in the Complex Model, vehicle testing programs need
to incorporate a number of measures to assure the relevance of
their results.  These measures must, at minimum, address  the
following  concerns:

     •    The emission effects being measured must be the result
          of the specific fuel modification for which benefits are
          being claimed.

     •    The emissions effects being measured,  including
          nonlinear or interactive effects, must be quantifiable
          across the entire set of vehicles of concern.

     •    The emission effects must meet the statistical
          significance and explanatory power criteria used to
          develop the Complex Model.

To address these concerns, the Agency has promulgated a set of
fuel specifications, fuel blending and measurement tolerances,
vehicle selection criteria, and data analysis processes to govern
vehicle testing programs undertaken to augment the Complex Model.
The remainder of this section explains and justifies these
requirements.  In addition, a summary and analysis of substantive
comments received by EPA regarding its prior vehicle testing
proposals is presented.

1.   Specifications for Fuels Required for Vehicle Testing
Programs

a.   Orthogonal fuel design

     The properties of the fuels tested in a vehicle testing
program limit the emission effects that can be determined from the
test program's effects.  Two fuels with different levels  of the
fuel property of interest must be tested to determine whether a
first-order  (i.e., linear) effect of the fuel property on

                                239

-------
emissions exists and to estimate the size of this effect.  Fuels
containing three different levels of the fuel property must be
tested to determine whether a second-order  (i.e., squared)  effect
exists and to estimate its size.  Additional fuels must be tested
to determine the presence of interactive effects.

     In the ideal case, test fuels would be orthogonal to one
another.  A pair of fuels is said to be orthogonal with respect to
a fuel parameter when the two fuels are identical in all respects
except for that fuel parameter.  As the number of properties being
investigated grows, the number of fuels required to construct an
orthogonal set of fuels increases exponentially.  If two separate
levels of each fuel property are to be tested and there are N such
properties, then the number of fuels required equals two raised to
the Nth power:  2N.   If three  separate  levels  of each  fuel  property
are to be tested in order to determine the existence and size of
second-order effects, then the number of fuels required increases
to 3N.   In either case,  the  number of  fuels  increases  rapidly as
the number of fuel properties increases.  Since the exhaust
emission Complex Model incorporates seven fuel properties for VOC,
NOx,  POM, and butadiene, eight properties for exhaust benzene, and
ten properties for acetaldehyde and formaldehyde, an ideal vehicle
test program would contain thousands of fuels.  Such test programs
would be prohibitively costly to conduct.

     Three factors reduce the number of fuels required for vehicle
testing to more manageable levels, however.   First, the number of
fuels required for VOC and NOx testing can be reduced if testing
is conducted to identify second-order interactive effects only.
In other words, by assuming that interactive effects will not take
the form  (A2  x B)  or (A x B  x  C)  where  A,  B,  and C are fuel
parameters, many fuels need not be tested.  This assumption
appears reasonable, since the existence of such effects would
imply that simultaneous interactions between three separate
molecules significantly affect emissions.  Such interactions are
far less likely to occur than pairwise interactions; in fact,
chemical reactions involving three separate molecules typically
are modeled as sequential pairwise reactions.  By eliminating
three-way interactions, a comprehensive test program would require
fewer than 50 fuels.

     Second,  the Complex Model and Complex Model database already
include much of the information that a comprehensive program would
provide.  Hence combining the test program results with the
Complex Model and Complex Model database permits the size of the
test program to be reduced still further.

     Third, the Complex Model does not include all possible
effects, despite having analyzed a large body of test data.  Since
the size of any one test program is likely to produce fewer data
than the existing Complex Model database contains, it is unlikely
that effects not found when analyzing the Complex Model database

                                240

-------
would be found in an individual vehicle testing program.  As a
result, some effects can be eliminated from consideration.  In
other cases, notably for toxics emissions, measurement
uncertainties are sufficiently large to make the identification
and accurate estimation of interactive effects difficult.  It
would be pointless to require testing to identify interactive
effects for exhaust toxics since such effects cannot be identified
with accuracy from test data.

     Given the preceding discussion, EPA has identified fuels
that, if tested in a well-designed and well-conducted vehicle
testing program,  can provide sufficient information to quantify
genuine emission effects with sufficient confidence to supplement
the Complex Model with the estimated emission effects.  Two
separate sets of fuels have been identified.  The first set of
fuels is designed to provide sufficient information to extend the
valid range of fuel parameters already included in the Complex
Model.  In this case, the test program would be designed to
quantify the emissions effects of existing Complex Model fuel
parameters at levels that have not been adequate tested.  The
second set of fuels is designed to provide sufficient information
to add a new fuel parameter to the Complex Model.


b.   Fuels to be tested when extending the range of the Complex
     Model

     As discussed in Section V of the preamble, at least three
fuels must be tested when extending the range of the Complex
Model.  These three "extension fuels" are specified in Tables IV-
57 and IV-58.  Table IV-57 specifies the level of the parameter
being extended; Table IV-58 specifies all other parameters for the
three extension fuels.

Table IV-57:   Level of Existing Complex Model Parameters Being
               Extended

Fuel Property       Extension         Extension   Extension
Being Extended      Fuel #1           Fuel #2     Fuel #3

Sulfur, ppm         Extension Level      80         450
Benzene, vol%       Extension Level       0.5         1.5
RVP, psi            Extension Level       6.7         8.0
E200, %             Extension Level      38          61
E300, %             Extension Level      78          92
Aromatics, vol%     Extension Level      20          45
Olefins, vol%       Extension Level       3.0        18
Oxygen, wt%         Extension Level       1.7         2.7
Octane, R+M/2       87.5                 87.5        87.5
                                241

-------
Table IV-58:
Levels for Fuel Parameters Other Than Those Being
Extended
Fuel Property

Sulfur, ppm
Benzene, vol%
RVP, psi
E200, %
E300, %
Aromatics, vol%
Olefins, vol%
Oxygen, wt%
Octane, R+M/2
     Extension
     Fuel #1

     150
       1.0
       7.5
      50
      85
      25
       9.0
       2.0
      87
Extension
Fuel #2

150
  1.0
  7.5
 50
 85
 25
  9.0
  2.0
 87
Extension
Fuel #3

150
  1.0
  7.5
 50
 85
 25
  9.0
  2.0
 87
     These fuels are designed to test the effects of the fuel
parameter being extended at three separate levels in order to
estimate the linear and squared effects of the parameter in
question.  Extension Fuel 1 will have levels of the parameter
being tested that are equal to the most extreme value for which
the augmentation will be valid.  Extension Fuels 2 and 3 will have
levels of the parameter being tested within the range of the
Complex Model,  which allows the test program to correlate its
results to those for the Complex Model.  The properties of these
two fuels are near the limits of the valid range of the model for
reformulated fuels but are sufficiently far within those limits so
as to assure the reliability of the Complex Model correlations  (as
discussed in Section D, the Complex Model correlations become less
reliable for fuels at or near the limits of the valid range of the
model) .

     The three fuels described above do not provide the
information needed to evaluate interactive effects.  However, EPA
does not anticipate that test programs of the size required to
augment the model would be sufficient to identify interactive
effects that were not included in the Complex Model, which is
based on a much larger set of data than would be available from
augmentation test programs.  Hence EPA does not require testing
for interactive effects unless such effects are already
incorporated in the Complex Model.  For example, the VOC model
includes interactive effects for aromatics (with E300) but does
not include interactive effects for sulfur.  As a result, a test
program to extend the range of sulfur levels covered by the model
would not be required to test for interactive effects, but a test
program to extend the aromatics range would be required to do so.

   To evaluate interactive effects, two additional fuels are
required.  These two fuels must have the following
characteristics:
                                242

-------
            The parameter being tested must be at the level
            specified for Extension Fuel 1, i.e., it must be at the
            most extreme value for which the augmentation will be
            valid.

            Other parameters which interact with the parameter being
            tested according to the Complex Model (for any
            pollutant) must be at two different levels in the two
            fuels.  The first fuel (Extension Fuel #4) must contain
            these other parameters at the level specified in Table
            IV-57 for Extension Fuel #2, i.e., at relatively low
            levels.  The second fuel (Extension Fuel #5)  must
            contain these other parameters at the level specified in
            Table IV-57 for Extension Fuel #3, i.e., at relatively
            high levels.

            All other parameters must be present at the intermediate
            levels specified in Table IV-58.
  c.   Fuels to be tested when adding a new parameter to the Complex
       Model

       More fuels must be tested when adding a new parameter to the
  Complex Model for several reasons.  First, the Complex Model
  database cannot be used to supplement the results from the vehicle
  testing program.  Second, the Complex Model cannot be used to
  reduce the testing needed to identify and quantify any interactive
  effects.  In order to quantify linear, squared, and interactive
  effects for the new parameter,  seven fuels must be tested
  (referred to as "Addition Fuels").  The specifications for these
  seven fuels are listed in Table IV-59.
Table IV-59:
Fuel Property
Properties of Fuels To Be Tested When Augmenting The
Model With A New Fuel Parameter

 	FUELS	
Sulfur, ppm     150    150
Benzene, vol%   1.0    1.0
RVP, psi        7.5    7.5
E200, %         50     50
E300, %         85     85
Aromatics, vol% 27     27
Olefins, vol%   9.0    9.0
Oxygen, wt%     2.1    2.1
Octane, (R+M)/2 87     87
New Parameter*   C      C+B
                        2
150
1.0
7.5
50
85
27
9.0
2 .1
87
B
35
0.5
6.5
62
92
20
2.0
2 .7
87
C
35
0.5
6.5
62
92
20
2.0
2 .7
87
B
500
1.3
8.1
37
79
45
18
1.5
87
C
500
1.3
8.1
37
79
45
18
1.5
87
B
                                  243

-------
C = Candidate level,   B = Baseline level


     Addition Fuels 1, 2, and 3 are necessary to determine the
 linear and squared effects for the parameter being tested.  Since
 emissions performance of reformulated fuels is calculated
 relative to  the emissions performance of Clean Air Act baseline
 gasoline, Addition Fuel #3 must contain this parameter at a level
 representative of that found in Clean Air Act baseline gasoline
 in order to  assign proper emissions credit for changes in this
 fuel parameter.  The parameter also must be tested at its
 "candidate"  level, the most extreme level at wich it is to be
 tested  (which can be higher or lower than its level in Clean Air
 Act baseline fuel), as specified  for Addition Fuel #1.   The
 parameter also must be tested at  a level midway between these
 first two levels, as in Addition  Fuel #2, in order to estimate
 the parameter's squared effects on emissions.  Squared effects
 are most easily determined if the parameter levels for the three
 fuels are evenly spaced.  For all three fuels, other fuel
 parameters must be present at levels which satisfy two
 requirements:

     •    They must be at levels  typical of reformulated fuels
          that are likely to be produced during 1995-1999.

     •    They must be at levels  for which the Complex Model
          correlations are highly reliable, i.e., they must be at
          levels near the center  of their range in the Complex
          Model database.

     Addition Fuels 4 through 7 are necessary to identify and
 quantify any interactive effects  associated with the parameter.
 Addition Fuels 4 and 5 represent  "clean" fuels in that all
 parameters other than the one being tested are set at levels that
 generally result in emission reductions from baseline levels.
 Addition Fuels 6 and 7 represent  "dirty" fuels since all
 parameters other than the one being tested are set at levels that
 generally result in emission increases from baseline levels.  The
 "clean" and  "dirty" levels of these parameters are as widely
 separated as possible while still remaining within the data core
 and within the range expected to  occur in reformulated gasolines
 during 1995-1999.  Oxygen levels were set so as to be compatible
 with the use of any oxygenate under the "substantially similar"
 requirements of existing law.  The two "clean" fuels are required
 to contain levels of the parameter being tested that are equal to
 those specified for Addition Fuels 1 and 3, respectively; the
 same is true for the two "dirty"  fuels.  The wide separation in
 parameter levels among Addition Fuels 4 through 7 is designed to
 improve the  ability of the analysis method (discussed in Section
 G.4) to isolate and quantify interactive effects associated with
 the parameter being tested.
                               244

-------
2.    Blending and Measurement Tolerances

a.    Blending precision

     In the course of its own vehicle testing programs,  EPA has
found that there are limits to the precision to which fuels can
be blended, particularly when fuels are blended from
representative refinery streams.  These limitations appear to be
the result of limitations in the design and process control
capabilities of modern refineries.  EPA's experience, combined
with comments received from refiners and fuel blenders,  indicates
that fuel blenders cannot achieve precisions beyond those
specified in Table IV-60.  In recognition of these limitations,
EPA has required that test fuels blended to the specifications
outlined in Section G.I fall within these tolerances of the
specified values.


Table IV-60:     Fuel Parameter Blending Tolerances

Fuel parameter                Blending tolerance

Sulfur content                ±25  ppm
Benzene content               +  0.2 vol %
RVP                           +  0.2 psi
E200 level                    ±  2 %
E300 level                    ±  4 %
Oxygenate content             ±  1.0 vol %
Aromatics content             ±  2.7 vol %
Olefins content               ±  2.5 vol %
Saturates content             ±  2.0 vol %
Octane                        ±  0.5
Detergent Control Additives   ± 10% of the level required by
                              EPA's detergents rule
Candidate parameter           To be determined as part of the
                              augmentation process


b.    Impact of blending variation

     Variation in test fuel properties from those specified in
Section G.I will tend to reduce the statistical power of the test
program.  Such variations would make estimates of emission
effects using simple averages of percent change in emissions per
unit change in each fuel parameter impossible.  However, the
analysis method outlined in Section G.4 can compensate for small
deviations from the test fuel specifications by utilizing the
existing Complex Model database to help disentangle the effects
of fuel modifications from one another.
c.   Measurement precision

                               245

-------
     Any measurement is subject to some degree of uncertainty.
This uncertainty can be separated into two types.  Uncertainty
can result from variations in separate measurements when using
the same measurement technique and instruments.  Uncertainty can
also result from variations in measured values when using
different measurement techniques or instruments.  The first type
of uncertainty can be reduced by taking multiple measurements; as
the number of measurements increases,  the average measurement
becomes an increasingly precise estimate of the "true" value that
would be measured if the measurement technique and instruments
were applied without error.  This type of uncertainty is
characterized by its repeatability level, which refers to the
probability that two successive measurements of the same sample
will fall within a certain range a specified percentage of the
time,  typically 95 percent.  The second type of uncertainty can
be reduced by calibrating the measurement technique and
instruments being used against a set of known values determined
through other, more reliable means.  This type of uncertainty
often is characterized by its reproducibility level, which refers
to the probability that measurements of the same sample at two
different facilities will fall within a certain range a specified
percentage of the time.

     Both sorts of measurement precision can affect the estimates
of emission effects from vehicle testing programs.  Inaccuracies
in estimates of fuel parameter levels or emission levels can
cause the analysis method outlined in Section G.4 to provide
inaccurate estimates of emission effects.  The procedures
outlined in the Code of Federal Regulations and §80.56-80.57 of
the reformulated gasoline regulations are designed to minimize
emission measurement errors, and the procedures outlined in
§80.46 are intended to minimize fuel property measurement errors.
Nonetheless, some measurement uncertainty is inevitable.

     EPA recognizes that for some fuel parameters, the size of
the measurement uncertainty can result in substantial uncertainty
in emission estimates.  Reducing this uncertainty is particularly
important in vehicle testing programs since any inaccuracies in
fuel parameter measurements can affect the resulting augmentation
to the Complex Model.  Inaccuracies in the augmentation, in turn,
can affect estimates of emission benefits for a large amount of
fuel and can influence capital investments by refiners and fuel
blenders.  As a result, EPA has concluded that the modest costs
associated with repeat testing of test fuel properties are more
than offset by the gain in modeling accuracy.  Hence EPA has
required test programs to measure the properties of their test
fuels to the accuracy specified in Table IV-61.  Test programs
must conduct repeat measurements of the fuel properties such that
the 95% confidence limits of the measured values for each
parameter (based on a standard t-test) about the mean measured
values are no larger than the tolerances presented in Table IV-
61.  Given the repeatability levels for relevant measurement

                               246

-------
methods, EPA does not expect a large number of repeat
measurements to be required.  The measurement tolerances,
repeatability levels for approved ASTM or EPA measurement
methods, and expected number of measurements required for each
parameter are presented in Table IV-61 (measurement methods refer
to ASTM designations unless specified otherwise).   Test programs
are responsible for providing documentation regarding the
repeatability, reproducibility, and correlation of their
measurement techniques as part of their augmentation application.
Table IV-61
Parameter
Fuel Parameter Measurement Tolerances for Fuel
Certification by Vehicle Testing
       Measurement
       Tolerance
             Method  &  95%
             Repeatability
API Gravity, °API
Sulfur, ppm
Benzene, vol%
RVP, psi
Octane, (R+M)/2
E200, vol%
E300, vol%
Oxygenate, vol%
Aromatics, vol%
Olefins, vol%
Saturates, vol%
Candidate parameter
                              0.1)
                              10)
                              0.03)
                              0.16)
                              0.1)
±  0.2       D1288  (±
± 10         D2622  (±
±  0.02      D3606  (±
±  0.05      D5191  (±
±  0.2       D4814  (±
±  2         D86    (±   3)
±  2         D86    (±   1.3)
±  0.2       GC/OFID
±  0.5       GC/MS
±  0.3       GC/MS
±  1.0
To be determined as part of the
augmentation process
No. of
Repeats

 1
 1
 3
11
 1
 3
 1
See text
See text
See text
See text
     In Table IV-61, no repeatability levels are provided for
oxygenate, aromatics, olefins, or saturates.  EPA anticipates
that oxygenate will be added in measured amounts to a fuel sample
of known volume to produce the required test fuels; since the
volume of oxygenate to be added can be controlled to extremely
high levels of precision and accuracy, EPA does not anticipate
that meeting the required tolerances will present petitioners
with significant difficulties.  The D86 measurement method does
not directly yield E200 and E300 values, but it can be modified
to do so by setting the sample temperature and measuring the
fraction of the sample that is recovered.  Alternatively, the D86
method can be used to derive T50 and T90 values, which can then
be converted to E200 and E300 values using the correlations
presented in Section A.  The repeatability levels presented in
Table IV-61 are based on this indirect method for determining
E200 and E300 values.

     The fuel measurement methods promulgated in the reformulated
gasoline rule include the use of the GC/OFID measurement method
for oxygenates and a gas chromatograph/mass spectrometer
measurement method for olefins and aromatics; saturates can be
                               247

-------
estimated by simple subtraction.  These methods have been
promulgated since the standard ASTM measurement methods yield
unacceptably high repeatability and reproducibility errors.  For
example, ASTM method D4815 for oxygenates yields 95%
repeatability levels of ± 0.83 volume percent for ethanol and ±
1.6 volume percent for MTBE.   Such methods would require 18
repeat measurements of ethanol blends on average and 64 repeat
measurements of MTBE blends on average.  The GC/OFID method has
95% repeatability levels of approximately  ± 0.24 volume percent
ethanol and  ± 0.37 volume percent MTBE.  Hence test programs
employing the GC/OFID method would need to conduct between two
and four repeat measurements for oxygenate concentration.
Aromatic and olefin measurements using ASTM method D1319 have 95%
repeatability levels of approximately 1.4 volume percent and 1.2
volume percent, respectively, and hence would require 8 and 16
repeat measurements, respectively.  EPA believes that the
measurement methods promulgated in the final RFC rule for
aromatics and olefins are capable of significantlyg better
repeatability.  Similarly, EPA's in-house experience with the RVP
measurement method  (D5191) indicates that repeatability levels of
0.05 psi are achievable, which would reduce the number of repeat
tests required from eleven to one.

     In addition to the requirements outlined in Table IV-61, EPA
requires that detergent control additive concentrations be
measured to within 2 percent of the level required by relevant
EPA regulations.  Since detergent control additives are expected
to be added to gasolines in small quantities after blending, EPA
believes that test programs will be able to control the
concentration of such additives to high levels of precision
without incurring additional cost.

     The required measurement accuracy for test fuel properties
is considerably more stringent than the accuracy required for
routine testing of reformulated gasoline in the field.  EPA
believes this greater stringency is warranted.  The
specifications for test fuels are carefully designed to provide
the data needed to evaluate proposed augmentations.  Variation in
the measured properties of the test fuels would affect the
results of an augmentation study.  Since these results could
influence fuel production decisions for a number of refiners, EPA
considers it desirable to eliminate potential sources of error
wherever practicable.

     The cost of greater fuel property measurement precision is
small compared with the total cost of vehicle testing programs.
EPA estimates that a typical vehicle test program of the size
required to augment the Complex Model could cost on the order of
$300,000 to as much as $1 million, depending on the cost of
procuring the necessary vehicles; the necessary fuel parameter
measurements would cost less than $10,000 (based on typical
commercial laboratory fee schedules).  Such costs are also small

                               248

-------
when compared with the potential savings to refiners resulting
from augmentations to the model.  In addition,  EPA expects the
accuracy and precision of existing measurement methods to
improve, since the reformulated gasoline program introduces
strong financial incentives to do so in order to recoup
compliance margins and reduce the risk of noncompliance.   For
these reasons, EPA has decided that the modest costs of greater
fuel property measurement accuracy are more than offset by the
resulting improvements in the accuracy of augmentations to the
model.

     However, the Agency does not consider such increased
accuracy to be practicable or cost-effective for routine testing
of commercial reformulated gasolines.  Repeat testing would be
required on each batch of gasoline, thereby increasing the cost
of reformulated fuels.  This cost increase would be particularly
onerous for low-volume gasoline marketers, since the increased
costs of fuel property testing would be spread across a
relatively small volume of fuel.  In addition,  the uncertainty
associated with single measurements of fuel properties will tend
to average out across multiple fuel batches produced or sold
commercially; such averaging is neither possible nor desirable in
a vehicle testing program.

3.    Vehicle Fleet Composition

a.    Principles of Fleet Balancing

     As discussed in Section A, EPA has found that vehicles with
different emission characteristics and different emission control
technology respond differently to fuel modifications.  To capture
these effects adequately in a vehicle testing program, the
vehicles tested should reflect the emission characteristics and
technology distribution of the in-use fleet.

     The purpose of the Complex Model is to predict the effect on
emissions of fuel modifications.  To assure that sufficient
information will be available to estimate fuel effects on
emissions, EPA has defined two guiding principles for vehicle
selection:

     1)   The proportion of vehicles of each vehicle type
          included in a test program should reflect the
          contribution of that vehicle type to in-use emissions.

     2)   The vehicles included in a test program should include
          as wide a variety of emission classes and technology
          groups as possible without violating the first
          principle.

EPA used these guiding principles to determine the desired
composition of vehicle testing fleets.

                               249

-------
b.   Distribution by Emitter Class

     As discussed in Section III.B.4,  normal emitters account for
52% of exhaust VOC and exhaust toxics emissions in Phase I and
44.4% of these emissions in Phase II.   In keeping with Guiding
Principle (1) ,  these proportions suggest that approximately half
of the vehicles tested should be normal emitters and half should
be higher emitters.  By contrast, normal emitters account for 82%
of NOx emissions in Phase I and 73.8% of NOx emissions in Phase
II, which suggests that 75-80% of all vehicles tested should be
normal emitters.

     However,  the relative scarcity of higher emitter data
implies that the effects of fuel modifications on higher emitters
is known with less certainty than for normal emitters.  To
compensate for this greater uncertainty, EPA considered requiring
a disproportionate share of the vehicles tested to be higher
emitters.  At the same time,  EPA was reluctant to deviate too far
from the emission ratios presented above.  Hence the Agency
decided to require that half of all vehicles tested be higher
emitters.  This distribution assures that sufficient data on
higher emitters will be available to estimate the effects of fuel
modifications with confidence.  The Agency considers the increase
in the accuracy of emission estimates for higher emitters (for
VOC,  NOx, and toxics)  to more than compensate for the reduction
in the accuracy of normal emitter NOx emission estimates that
would result from requiring the proportion of normal emitters to
match the contribution of normal emitters to NOx emissions.   A
more thorough discussion of this issue is contained in the April
1992 NPRM.
c.    Technology groups

     As discussed in Section A.3,  EPA has modeled the response of
normal emitters by grouping vehicles with different emission
control technologies and allowing terms specific to each group to
enter the regression equations.  This approach was required by
the imbalance between the contribution of each technology group
to in-use normal emitter emissions and the proportion of test
data available for each technology group.  If the proportion of
test data were to match the proportion of emissions due to each
technology group, then the population of test vehicles would
reflect the origin of in-use emissions and separate models for
individual technology groups would be unnecessary.

     EPA has chosen to reduce the burden of vehicle testing by
eliminating the need for technology group-specific regression
terms.  Instead, EPA has chosen to require organizations involved
in vehicle testing to recruit a balanced set of normal emitting
vehicles according to Principles (1) and (2).   As a result,

                               250

-------
separate models for each technology group need not be constructed
when augmenting the model through vehicle testing.

     To simplify the process of vehicle specification for the
normal emitter test fleet,  EPA has developed a schedule to guide
vehicle acquisition.  The schedule identifies the vehicle
characteristics required of the first, second, third, etc.
vehicle.  An organization conducting vehicle testing is required
to include at least ten normal emitting vehicles to assure the
statistical validity and reliability of their results; such
organizations have the option of testing additional vehicles to
increase the statistical power of their test program.  Additional
vehicles are to be added in the order specified so as to maintain
a balanced test fleet.  The normal emitting vehicle acquisition
schedule is presented below in Table IV-62.
                               251

-------
Table IV-62:   Normal Emitting Vehicle Acquisition Schedule
 Veh.  Fuel
  #  System
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Multi
Multi
TBI
Multi
Multi
Multi
TBI
ulti
TBI
Multi
Multi
Multi
Carb
TBI
Multi
Multi
Multi
TBI
Multi
Multi
Catalyst

 3W
 3W
 3W
 3W+OX
 3W
 3W
 3W
 3W+OX
 3W+OX
 3W
 3W
 3W
 3W+OX
 3W
 3W+OX
 3W
 3W
 3W
 3W+OX
 3W
Air
Injection

No Air
No Air
No Air
Air
No Air
No Air
No Air
Air
Air
Air
No Air
No Air
Air
No Air
Air
No Air
No Air
No Air
Air
No Air
EGR
Tech.
Group
EGR
NoEGR
EGR
EGR
EGR
NoEGR
EGR
EGR
EGR
EGR
EGR
NoEGR
EGR
EGR
EGR
EGR
NoEGR
EGR
EGR
EGR
1
2
3
4
1
2
3
4
7
5
1
2
9
3
4
1
2
3
4
1
Manu-
facturer

GM
Ford
GM
Ford
Honda
GM
Chrysler
GM
Chrysler
Toyota
Ford
Chrysler
Toyota
Ford
GM
Toyota
Mazda
GM
Ford
Nissan
Legend:
     Fuel system:
          Multi
          TBI
          Carb

     Catalyst:
          3W
          3W+OX  =
    Multi-point  fuel  injection
    Throttle  body fuel  injection
     Carburetted
       3-Way catalyst
       3-Way catalyst  plus  an  oxidation  catalyst
     Air Injection:
          Air    =  Air injection
          No Air =  No air injection
     EGR:
          EGR    =  Exhaust gas recirculation
          No EGR =  No exhaust gas recirculation
                               252

-------
Table IV-63:  Technology Group Definitions

Tech      Fuel                      Air
Group     System       Catalyst     Injection     EGR
  I       Multi        3W           No Air        EGR
  2       Multi        3W           No Air        No EGR
  3       TBI          3W           No Air        EGR
  4       Multi        3W+OX        Air           EGR
  5       Multi        3W           Air           EGR
  6       TBI          3W           Air           EGR
  7       TBI          3W+OX        Air           EGR
  8       TBI          3W           No Air        No EGR
  9       Garb         3W+Ox        Air           EGR
     EPA has not specified a similar schedule for higher emitters
since the Agency does not believe technology group distinctions
to affect the emissions response of such vehicles to fuel
modifications.  This issue is discussed more fully in Section
A. 2 .

4.   Analysis Process

     The analysis process promulgated in the reformulated
gasoline rule is designed to extract the maximum amount of
information from the limited data available from each vehicle
testing program.  It incorporates statistical analysis techniques
analogous to those used to develop the Complex Model, and it
utilizes the existing Complex Model database where possible to
supplement the data from individual test programs.  The analysis
process is designed to achieve the following goals:

     •    Accurate estimation of the linear and nonlinear
          emission effects due to the fuel property being tested.

     •    Accurate estimation of interactive effects between the
          fuel property being tested and other fuel properties
          known to affect emissions.

     •    Isolation of the effects of the fuel property being
          tested on emissions from normal and higher emitting
          vehicles.

     •    Isolation of the effects of the property being tested
          from the effects of related or correlated fuel
          properties already included in the model.

     •    Preservation of already-established Complex Model
          effects within the data core.

     The remainder of this section discusses the rationale for
the analysis method and statistical techniques promulgated in the

                               253

-------
reformulated gasoline rule.  It also discusses the basis for the
differences in analysis process when testing extension or
addition fuels.
a.   Analysis method and statistical techniques

     The analysis methods to be used to analyze test program data
are similar to those used to develop the Complex Model.  The
basic process involves the following elements; many of these
techniques are discussed and justified in greater detail in
Section A.

     1.    Separate analyses are to be conducted for normal and
          higher emitters, since past experience has shown that
          emission responses to fuel modifications differ for
          these two types of vehicles.  In addition, petitioners
          must demonstrate that their higher emitting vehicles
          were stable over the course of the test program in
          terms of their emissions on a given fuel.  Past EPA
          testing experience has shown that emissions from higher
          emitters often demonstrate time trends or variations
          that are unrelated to and considerably larger than the
          effects of fuel modifications; control of such
          variability is necessary if genuine fuel effects are to
          be identified.

     2.    All models are to fit polynomials (first-order for
          toxics, no higher than second-order for VOC and NOx) to
          the natural logarithm of emissions,  as was done for the
          existing Complex Model.  EPA and others have found that
          transforming emissions using the natural logarithm
          results in improved model accuracy and more normally
          distributed residuals.

     3.    Petitioners must demonstrate that collinearity problems
          are not exceedingly large.  Variance inflation values
          in excess of 10 will be interpreted to indicate that
          the models constructed by the petitioner contains
          severe collinearity problems, in keeping with standard
          statistical practice.

     4.    Petitioners must demonstrate that the residuals from
          their regressions are approximately normally
          distributed.  The statistical analysis process outlined
          in the regulations requires such a distribution.
          Models without normally-distributed residuals cannot be
          generated or evaluated using the more powerful
          techniques associated with normally-distributed data.

     5.    Petitioners must demonstrate that their analysis has
          adequately balanced potential overfitting and

                               254

-------
          underfitting of the data through the use of Mallows'  Cp
          criterion or other appropriate methods.  Underfitted
          models have too few terms to explain the observed
          variation in the data; overfitted models have too many
          terms, some of which are likely to be erroneous.

     6.   All regressions are to include separate terms for each
          vehicle-study combination in the form of dummy
          variables, as was done for the existing Complex Model.

     7.   To retain consistency with the existing Complex Model,
          all regressions must retain terms with significance
          levels of 0.90 or higher.


b.   Special Analysis Requirements When Extending Existing Model
     Parameter Ranges

     Test programs designed to extend the valid range of the
model for existing Complex Model parameters can utilize the data
already available in the Complex Model database.  This additional
data increases the statistical power of the analysis process and
allows the augmentation to be integrated with the existing
Complex Model smoothly.  The analysis process involves the
following unique elements:

     1.   The test data on all extension fuels tested must be
          combined with the Complex Model database used to
          develop the Complex Model being promulgated today.  All
          analyses must be performed on this combined database.

     2.   The first step in analyzing this combined data set is
          to fit a standard regression model to the combined
          database.  The model must contain separate terms for
          each vehicle-study combination.  It must also contain
          all terms present in the existing Complex Model.  The
          coefficients for those terms which do not involve the
          parameter being extended must be set equal to their
          values in the existing centered, consolidated Complex
          Model equations.  The coefficients for terms which do
          involve the parameter being extended  (including any
          linear, squared, and interactive terms) are to be
          estimated by the regression.  This approach assures
          that the only changes in the existing Complex Model
          will occur for those terms involving the parameter
          being extended.

     3.   In some cases, the existing consolidated Complex Model
          may not contain a squared term for the parameter being
          extended.  However, the additional data made available
          through the test program may warrant inclusion of such
          a term.  To allow such terms to be included in the

                               255

-------
     model,  a stepwise regression procedure should be used
     to determine whether the squared term for the parameter
     being extended would be statistically significant.   The
     linear term must be retained even if it is no longer
     significant once the second-order term is added.

4.   The fuel terms included in the models described above
     must be expressed in centered form.   The centering
     value for these terms must be identical to the
     centering value used to develop the Complex Model being
     promulgated today.   Centering values can be found in
     Section A.8 and will be made available to petitioners
     by EPA upon request.  These centering values are
     necessary during development of the augmented model;
     the same centering values as were used during
     development of the current Complex Model must be used
     since the coefficients from that model will be used for
     a number of terms in the models described previously.
     Once the augmented model has been developed, it should
     be uncentered using the process described in Section
     A.8.

5.   The model resulting from the foregoing steps is the
     augmented model.  It can be used for levels of the fuel
     parameter in question only when it is present in fuels
     at levels between the extension level and the data
     core.  The performance of such fuels is to be
     calculated as follows, with all performances being
     expressed as percentage changes in emissions from
     baseline levels:

     a.    Up  to the limits of the data core, the
           unaugmented Complex Model must be used to
           determine the percentage change in emissions due
           to  the parameter in question.  In other words,
           the unaugmented model is to be used to estimate
           the performance of a fuel identical to the
           target fuel, but with levels of the augmentation
           parameter set equal to the nearest limit of the
           data core.

     b.    The performance of the fuel described in step
            (5.a) must be determined using the augmented
           model.

     c.    The performance of the actual target  fuel must
           be  determined using the augmented model.

     d.    The performance of the actual target  fuel, in
           percentage terms, shall be set equal  to the
           following value:
                          256

-------
                100 -  [(100 + a) *  (100 + c)/(100 + b)]

          These calculations maintain equivalency with the
Complex Model up to level at which the existing Complex Model
database becomes so sparse as to make the existing Complex Model
equations less reliable.  This equivalency is necessary to avoid
fungibility problems from being introduced by model
augmentations.  Without the adjustments described above,  a model
augmentation might give greater credit for a fuel modification at
a given parameter level than the existing Complex Model.   As a
result, mixing of fuels certified using an unadjusted augmented
model and the existing unaugmented model could produce a mixture
in which each original constituent met the emission reduction
requirements of the program but the mixture itself would not.
This situation, and the compliance and certification problems it
presents, can be avoided by using the adjustments described
above.
c.   Special Analysis Requirements When Adding New Model
     Parameters

     Test programs designed to augment the model with a new
parameter require a somewhat different analysis process involving
the following elements:

     1.   The first step is to adjust the measured emission
          levels to account for deviations in fuel
          characteristics other than the one being tested from
          those specified.  This process involves using the
          Complex Model to determine the percentage change in
          emissions between the desired and actual fuels,  then
          adjusting the measured emissions accordingly.

     2.   The second step is to develop a model based on this
          adjusted emissions data from addition fuels 1, 2, and
          3.  This model must include separate dummy variable
          terms for each vehicle, a linear term for the parameter
          in question,  and a squared term for the parameter in
          question.  If the squared term is not significant,  it
          must be dropped from the model, which must then be re-
          estimated.  If the linear term in the resulting model
          is not significant, then it must be dropped from the
          model.  In such cases, EPA will not grant the
          augmentation and the analysis process (and test
          program) can end.  However, the analysis results still
          must be reported to EPA, since these results may be
          helpful in revising the Complex Model or evaluating
          other augmentation petitions.

     3.   The third step is to build a stepwise regression on a
          data set which combines the test data from all seven

                               257

-------
          fuels and the data in the existing Complex Model
          database.  Using this data set, the vehicle terms for
          the test fleet must be re-estimated.  The initial model
          must include separate terms for each vehicle-study
          combination; it must also include all terms present in
          the centered, consolidated Complex Model along with
          their coefficients, and it must include the linear and
          squared terms (if statistically significant)  and their
          coefficients as estimated previously based on addition
          fuels 1, 2,  and 3.  Interactive terms between the
          parameter in question and existing Complex Model fuel
          parameters should be added in a stepwise fashion,
          subject to the conditions discussed in Section G.4.a.
          Once the forward-fit model is complete, a backwards
          elimination process can be used to satisfy Mallows' Cp
          criterion.  However, only the interactive terms added
          in the stepwise regression process may be deleted in
          the backwards elimination process.  This step assures
          the equivalency of the augmented and unaugmented
          Complex Models within the range of the unaugmented
          model, which is necessary to avoid fungibility problems
          as discussed in Section G.4.b.

     5.   The fuel terms included in the models described above
          must be expressed in centered form.  The centering
          value for these terms must be identical to the
          centering value used to develop the Complex Model being
          promulgated today.  Centering values can be found in
          Section A.8 and will be made available to petitioners
          by EPA upon request.  These centering values are
          necessary during development of the augmented model;
          the same centering values as were used during
          development of the current Complex Model must be used
          since the coefficients from that model will be used for
          a number of terms in the models described previously.
          Once the augmented model has been developed,  it should
          be uncentered using the process described in Section
          A.8.

     6.   The model resulting from the foregoing steps is the
          augmented model.  It must be used to calculate the
          performance of all fuels which claim benefits from the
          parameter in question.

5.   Response to Comments

     EPA has received a number of comments in response to the
April 1992 and February 1993 proposals.  The Agency has
incorporated such comments into the final regulations governing
vehicle testing where such comments were appropriate or
adequately supported by data.  However, these regulations do not
incorporate all the comments received by EPA.  What follows is a

                               258

-------
description of the major comments received by EPA,  the Agency's
response to these comments,  and the Agency's reasons for not
incorporating those comments which have not been addressed in the
final rule.  EPA also received a number of comments which are not
relevant to this rulemaking; such comments are not discussed
below.   Comments have been grouped by topic to simplify matters.


a.   Direct Certification

     Several commenters requested that vehicle testing be
permitted as a direct alternative for fuels certification instead
of requiring augmentation of the Complex Model.  As discussed in
the preamble and in the introduction to this section,  however,
EPA believes that direct fuel certification through vehicle
testing is neither practicable nor desirable.  Certification of a
specific fuel would not extent to fuels with minor differences in
composition or other fuel properties, yet such differences
inevitably arise in the course of fuel production.   As discussed
in Section G.2,  fuel producers cannot precisely control the
properties of their fuel, which would make production of a
specific fuel formulation essentially impossible.  Direct fuel
certification through testing is also not desirable since, as
discussed in the preamble and in Section A, the results of a
single test program are less reliable and more vulnerable to
accidental or deliberate manipulation than is a model based on
the results of multiple test programs.


b.   Updates to the Model

     EPA originally proposed to update the Complex Model every
five years to incorporate the most recently available data on
emission effects of fuel reformulations.  Several commenters
objected to such frequent updates, noting that changes in the
model could invalidate investments in refinery technology
undertaken to meet standards using the previous version of the
model.  Commenters called for updates no more often than once
every 10-15 years or asked that refiners be able to grandfather
fuel certification under an earlier version of the model for ten
years.  EPA has reconsidered its plans to update the model every
five years, in part because it is difficult to predict when
sufficient additional data will be available to warrant such an
update.  Instead, EPA plans to update the model through a formal
rulemaking process that will be undertaken when EPA determines
that sufficient new information is available to warrant such
action.  As part of this rulemaking process, EPA will consider
the impact of changes in the model on refiners' capital
investments.
c.   Test Fuels

                               259

-------
     EPA received numerous comments on its proposed test fuels.
Many of the comments received in response to the April 1992
proposal have been incorporated in the final regulations
governing vehicle testing.  For example,  one commenter suggested
that the "candidate-baseline" fuel defined in the April 1992
proposal should not specify that the level of the "candidate
parameter" be zero, since the parameter may be present at non-
zero levels in typical baseline fuels.  EPA agrees with this
comment and has modified its fuel requirements accordingly.
Another commenter objected to the test fuels required to augment
the simple model in the April proposal; EPA has since deleted the
simple model augmentation option from the rule.  A commenter
noted that standard deviations cannot be averaged across
pollutants, as EPA had proposed in its April 1992 proposal; the
final requirements for vehicle testing do not include averaging
of standard deviation or variance across pollutants.

     One commenter recommended that fuel parameter values for
interactive effects testing be no greater than the levels found
in industry average gasoline.  EPA considers such an approach
inappropriate for several reasons.  First, industry average
gasoline is not representative of expected reformulated fuels in
some respects, notably oxygen, RVP, aromatics, and  (in Phase II)
sulfur levels.  Second, interactive effects are more easily and
accurately estimated when testing for such effects extends over a
wide range of fuel parameters.  Third, EPA recognizes that the
Complex Model will be used to assure compliance of conventional
fuels with the anti-dumping program, and such fuels can have fuel
properties that are quite different from those of industry
average gasoline.  Hence testing only 1990 industry average
gasoline for interactive effects would make identification of
interactive effects more difficult and would not provide
sufficient data to estimate such effects for typical reformulated
or conventional gasolines.

     A number of comments were received on detergent additive
requirements for test fuels.  Commenters generally supported
requiring such additives in test fuels since they will be
required in reformulated gasolines.  Some commenters recommended
specific additives or specific levels of additives; however, EPA
does not consider it appropriate to provide such specificity in
this rule.  Instead, EPA has required test fuels to meet the
detergent additive requirements set forth in EPA's impending
detergent additives rule.  In a related comment, the Agency was
asked to give credit for fuel modifications that result in
reduced emissions after extended use.  EPA does not believe it
appropriate to do so beyond the extent shown in the required
vehicle test program.  Such benefits are unlikely to be achieved
in-use unless two conditions are met:  the fuel modifications are
used on all gasolines sold by the petitioner indefinitely, and
all of the petitioner's consumers exhibit perfect brand loyalty.
Absent a showing that such conditions exist and can be assured,

                               260

-------
EPA does not believe it appropriate to augment the model with
emission effects that occur only after extended use.  EPA
believes that such modifications are more appropriately addressed
by other fuels-related programs required by Congress, such as
deposit control additives.

     Comments were received recommending deletion of the API
gravity and end point specifications for test fuels.  EPA has
deleted these specifications for fuel blending since these
characteristics are largely determined by characteristics with
specified blending tolerances.  EPA has also eliminated the
requirement that end point be measured for all test fuels.  End
point is closely correlated with E300, which is specified; in
addition, end point is difficult to measure accurately.  However,
EPA has chosen to retain the API gravity measurement.  This
parameter is relatively easy to measure and is specified in the
Act for Clean Air Act baseline fuel.  The Agency believes that it
must be measured to assure comparability with Clean Air Act
baseline fuel and is not aware of any measurement difficulties
associated with it.  Commenters also asked the Agency to
establish a minimum octane level instead of a blending tolerance.
The Agency has decided to retain its blending tolerance
requirement while adding a minimum octane requirement, thereby
limiting octane values to the range of 86.5 to 87.5.  Such tight
control is feasible, as discussed in Section G.2, and EPA
believes such control to be necessary to prevent distortion of
test results.  Octane is known to affect engine performance, per-
gallon fuel efficiency, and spontaneous ignition (engine knock),
all of which can affect emissions.

     One commenter suggested that test fuels be based on typical
at-the-pump fuels.  However, as discussed in Section G.I, test
fuel characteristics must be controlled tightly in order to
extract useful information from test program results.  At-the-
pump fuels do not allow such tight control.  In addition, EPA
believes that the requirement that test fuels be blended from
typical or realistic refinery streams assures that test fuels
will have characteristics similar to those of typical commercial
gasolines, except for those deliberately altered as part of the
test program design.


d.   Measurement methods and data requirements

     EPA received several comments on its proposed measurement
methods and data requirements to support augmentation requests.
Commenters suggested that EPA permit the use of equivalent
testing methods and notify petitioners of data requirements
before testing begins.  EPA believes that these comments are
adequately addressed by the requirement that petitioners obtain
approval for their proposed test program before beginning
testing.  This requirement provides petitioners with a means of

                               261

-------
requesting and obtaining approval for the use of equivalent
testing methods.  It also allows EPA to identify data
inadequacies in proposed test programs before testing begins.

     Several commenters recommended that EPA base its fuel
property measurement methods on standard ASTM methods.   EPA has
promulgated a set of approved measurement methods which match
standard ASTM practices in most cases.  However, EPA found it
necessary to promulgate alternative methods in some cases in
order to assure sufficient accuracy and precision to meet the
needs of the program.  EPA plans to submit these alternative
methods to ASTM for approval.  Since such approval can take
several years,  however, EPA believes it necessary to proceed with
the program without ASTM approval for these alternative methods.
Interested parties are also permitted to use their own preferred
measurement methods if they can demonstrate, to the Agency's
satisfaction, that these methods provide comparable accuracy to
those promulgated by EPA.  In the case of vehicle testing, the
Agency recognizes that the required measurement precision may be
beyond the capability of a single test using either standard ASTM
methods or the methods promulgated by EPA.  Such precision and
accuracy can be achieved through multiple fuel property
measurements or the use of more precise ASTM methods than are
standard.  As discussed in Section G.2, the modest cost of
replicate fuel measurements or more precise measurement methods
is more than justified by the gain in modeling accuracy.

     Commenters also objected to the requirement that carbon
dioxide be measured during vehicle testing.  EPA considers this
information necessary to assure the quality of the test program.
In particular,  carbon dioxide data is necessary to evaluate the
fuel economy performance of vehicles being tested; unexpectedly
low fuel economy provides an indication of vehicle malfunction or
improper test procedures.


e.   Vehicle Selection

     Several commenters suggested that EPA allow test programs to
select vehicles at random from the in-use fleet; other commenters
recommended basing the distribution of vehicles on 1990 market
share alone.  However, as discussed in Section G.3, the test
fleet composition must reflect the contribution of each type of
vehicle to the total emissions inventory.  Selecting vehicles at
random from the in-use fleet would not achieve such composition
except by chance, and selecting vehicles based only on market
share would ignore differences in emission levels among vehicles.
Further, EPA is concerned that petitioners may restrict their
vehicle acquisition efforts to those vehicles expected to respond
strongly to the fuel modifications being tested.  EPA's test
fleet requirements limit petitioners' ability to manipulate the
outcome of test programs in this manner.  Finally, a test fleet

                               262

-------
based only on market share or based on random selection from the
in-use fleet would be unlikely to result in a sufficient number
of higher emitters for two reasons.  First, 1990 technology
vehicles currently on the road are less likely to be higher
emitters than would be the case if the entire fleet consisted of
such vehicles, as is necessary to estimate the in-use impact of
gasoline formulations on emissions.  Second,  EPA has found that
owners of higher emitting vehicles are less willing to allow
their vehicles to be tested than are owners of normal emitters.
For these reasons, EPA has decided to retain its test fleet
composition requirements instead of permitting random selection
of test fleets from the in-use fleet or basing its test fleet
requirements only on market share.  EPA believes these
requirements assure an appropriate mix of technology types,
models, emitter classes, and engine displacements.

     A number of commenters objected to the test fleet
requirements regarding higher emitters proposed in the April 1992
NPRM.  EPA believes its current proposal incorporates these
comments where valid.  As discussed in Section G.3, higher
emitters must be tested in order to evaluate the actual in-use
emission effects of fuel parameters.  Such vehicles account for a
large fraction of in-use  emissions, and the Complex Model
indicates that such vehicles respond differently to fuel
modifications than do normal emitters.  This conclusion is
strengthened by the fact that each interim version of the Complex
Model showed higher emitters to respond differently than normal
emitters to fuel modifications.  In response to comments and its
own revised analysis of higher emitters, EPA has eliminated the
requirement that higher emitters match the vehicle technology
distribution set forth in Table IV-62.  As suggested by several
commenters, EPA has attempted to classify higher emitters by
failure mode in order to assure that the higher emitter test
fleet is representative of in-use higher emitters.  However,
available data is not sufficient to allow such classification and
commenters did not submit such data to EPA.  EPA asked for and
received comment on a proposal to allow intentional disablement
of normal emitting vehicles' emission control systems to simulate
higher emitters and reduce the cost of procuring a representative
test fleet.  Most commenters opposed this option, arguing that
intentionally disabled vehicles would not be representative of
in-use higher emitters.  In addition, EPA was unable to determine
an appropriate basis for such disablements, asw discussed above.
One commenter expressed concern that higher emitters would be
much more difficult to locate once enhanced I/M programs are in
place.  However, EPA's analysis of its I/M Repair Database
suggests that approximately one-fourth of the in-use fleet would
consist of higher emitters when enhanced I/M is in place.
Further, 1990 technology vehicles currently on the road would
tend to be higher-mileage, higher-emission vehicles by the time
enhanced I/M programs in place, which should make acquisition of
higher emitters less burdensome.

                               263

-------
     Another commenter asked that EPA provide a list of
equivalent 1990 vehicles.  EPA has not been able to provide a
comprehensive list of such vehicles but has included in Appendix
IV-F a list of the vehicles in the Complex Model database deemed
to have 1990-equivalent technology.  Petitioners are encouraged
to coordinate their vehicle selection process with EPA and with
the vehicle manufacturers to assure that all included vehicles
qualify as 1990-equivalent technology vehicles.


f.   Analysis method

     Commenters expressed concern over a number of requirements
in EPA's previous proposals, such as the repeatability
requirement and the use of stepwise regression techniques.  The
final requirements for vehicle testing do not include
repeatability requirements except insofar as the vehicle's
emission characteristics must be shown to be stable when tested
on the same fuel.  The need for this requirement is discussed in
Section G.4 and Section A.  Stepwise regression techniques have
been found to be necessary during the development of the Complex
Model, as discussed in Section A.4; in fact, one of the
commenters who objected to stepwise regression techniques used
such techniques in developing their own complex model.   No
suitable alternative techniques were identified by commenters,
and EPA believes the analysis method required for augmentation
petitions sufficiently restricts the use of stepwise regression
techniques to prevent their mis-application.
     Another commenter asked that EPA drop the requirement that
emission changes be shown within the first 50 miles of operation
upon being acquired for vehicle testing.  This requirement has
been dropped, in part because the required preconditioning,
stabilization, and testing procedures may require greater than 50
miles of vehicle operation.
g.   Nonexhaust Testing

     EPA's April 1992 and February 1993 proposals allowed
augmentation of the nonexhaust model.   However,  EPA no longer
believes that such augmentations should be permitted for reasons
discussed in the preamble to this rulemaking.  As a result,
comments regarding nonexhaust emission testing procedures are no
longer relevant.
                               264

-------
H.   Appendices
Appendix IV-A:

Appendix IV-B:

Appendix IV-C:



Appendix IV-D:


Appendix IV-E:


Appendix IV-F:
The Random Balance Algorithm

Glossary and Nomenclature

MOBILE4.1 Input File Used to Develop the Simple
Model and the Phase I Complex Model (must be run
with interim version of MOBILE4.1)

MOBILE5a Input File Used To Develop the Phase II
Complex Model

SAS code used to develop "raw" exhaust emission
models

Included 1990 Technology Vehicles in the Complex
Model Database
                               265

-------
Appendix IV-A:  The Random Balance Algorithm


The random balance algorithm was developed by H.  T.  McAdams of
the Advanced Computing Center of Argenta, Illinois.   This
appendix is a user's manual developed by Dr.  McAdams for the
application of this algorithm to the development and
simplification of the Complex Model.   The mathematical
manipulations program that Dr.  McAdams uses in all his work is
Matlab, such that the algorithm described below is given in the
context of Matlab language and instructions.


The random balance algorithm is useful for two purposes:

l)To determine the ranking of terms in the Complex Model in order
of their contribution to reduction of the residual sum of squares

2)To consolidate equations for separate tech groups into one or
more equations yielding essentially the same outputs as the
unconsolidated model
I.   RANDOM NUMBER GENERATION

The procedure begins with the generation of a set of N "random
blends." In previous applications, N has been taken as 2000.
However, other values of N can be used, up to the RAM capacity of
the PC being used.

1.1 Linear Terms

The starting point in the random-balance process is generation of
a matrix of random numbers.  The matrix has N = 2000 rows and as
many columns are there are fuel variables  (linear terms only) in
the model.  For VOC, there are seven: oxygen, sulfur, RVP, E200,
E300,  aromatics and olefins.

NOTE:  It is important that the random numbers be drawn from a
uniform, not a normal, distribution.  In the ACCA procedure, the
random numbers are uniform on the interval[-0.5,  0.5], but other
conventions could be used provided subsequent computations are
programmed consistent with the choice of the interval.

Matlab generates random numbers on the interval [0,1] by the
simple instruction
             x = rand(2000,7)       (1)

This produces a 2000 x 7 matrix of numbers randomly distributed
between 0 and 1.  To transform to the interval [-0.5,0.5], it is
necessary only to subtract 0.5 from all values of the matrix.
In Matlab, this is done by the simple instruction

                               266

-------
x = x - 0.5
                                 (2)
Or, (1) and (2) can be combined to the single instruction

             x = rand(2000,7) - 0.5    (3)

It is well to check the resulting array for zero mean and unit
correlation matrix.  Such a check can be done quickly and simply
by the instructions

             meanx = mean(x)      (4)

           corrcoefx = corrcoef (x)     (5)

Meanx is a row vector for which all elements should be
approximately zero.   Corrcoefx is a 7 x 7 matrix in which all
elements on the main diagonal should be 1, and all off -diagonal
elements should be approximately 0.   Several types of plots can
be easily and quickly made to verify this, notably
plot (corrcoefx ( : ) )  or bar (corrcoefx ( : ) ) .
1.2 Quadratic and Interaction Terms

The Complex Model potentially contains linear, quadratic  and
first-order interaction terms.  To augment the above random
matrix, one needs to annex a set of columns representing squares
of  the variables (7 columns) and another set of columns
representing products of all variables taken two at a time  (21
columns).

The square terms can be generated simply by the instruction x.*x.
The dot after the first x causes each column to be multiplied  by
itself row by row, rather than as an inner product.

A simple .m-file to generate all the interactions is  intergen.m
(see below).   It can be easily programmed in another language  by
simply reproducing the loops shown.

                           Table  IV-A-1
                     INTERGEN.M
                     function y=intergen(x)
                     [r,c]=size(x);
                     for i=l:c,
                     for j=i+l:c,
                     q=x(:,i).*x(:,j);
                     y=[y q];
                     end
                     end
                               267

-------
The linear, square and interaction parts of the moment matrix can
be combined simply by writing

         ucentall =  [x x.*x intergen(x)]   (6)

The name ucentall is simply a label (anything else would do),  but
it identifies the result as "all terms of a centered uniform
distribution."

As before, one should check the means and correlations and
possibly visualize the results as a line or bar plot before
proceeding further.

NOTE: Equation (6) will be used in another connection later on.
2.   TRANSFORMATION TO REAL FUEL-VARIABLE QUANTITIES

The 2000 x 7 matrix generated above ranges from -0.5 to 0.5.   We
need to convert these random numbers to corresponding values of
the actual parameters.   For this purpose, the range [-0.5, 0.5]
has to be mapped into the corresponding actual range of the fuel
variables.

In previous application of the random-balance methodology, the
ranges of the seven variables were as follows:

                           Table  IV-A-2
Fuel variable
Oxygen
Sulfur
RVP
E200
E300
Aromatic s
Olef ins
Minimum
0
50
7
32
76
20
2
Maximum
3.7
350
10
64
93
45
20
Other values can be used if desired; however, note that it is
assumed that all values within the specified range are equally
likely.

NOTE: In the EPA July 14, 1993 submission to the docket, the
"Limits of Data in the Complex Model Database" are slightly
different, but the only notable departure is that the upper limit
for sulfur is given as 450 rather than 350.

Changing the range [-0.5, 0.5] to the tabled ranges of the fuel
variables is a simple matter of linear transformation:
                               268

-------
      y = x * (Max - Min) +  (Max + Min)/2    (7)

Thus, for oxygen, the transformation is

      y = x * (3.7 - 0.0) +  (3.7 + 0.0)/2 = 3.7 x + 1.85

Similarly, for sulfur,  the transformation is

     y = x * (350 - 50)  + (350 + 50)/2 = 300 x + 200

and so on.

The Matlab .m-file to perform this transformation is invtran4.m
and is listed below.

                          Table  IV-A-3
 INVTRAN4.M
    function y=invtran4(x)
    %where x is the 7-column matrix of scaled  fuel parameter
    % and y is the corresponding matrix of raw fuel-parameter
    %values
    yl=3.7*x(:,1)+1.85;
    %yl is oxygen
    y2=300*x(:,2)+200;
    %y2 is sulfur
    y3=3*x(:,3)+8.5;
    %y3 is RVP
    y4=32*x(:,4)+48;
    %y4 is E200
    y5=17*x(:,5)+84.5;
    %y5 is E300
    y6=25*x(:,6)+32.5;
    %y6 is aromatics
    y7=18*x(:,7)+11;
    %y7 is olefins
    y= [yl y2 y3 y4 y5 y6 y7] ;
    %Note: Terms are: oxygen sulfur rvp E200 E300 aromatics
    %olefins
    %The e200 transformation is based on a range from  32  to 64
    %as derived by regressing e200 on t50 and  evaluating  at 170
    %and 235.
    %The e300 transformation is based on a range from  76  to 93
    %as derived by regressing e300 on t90 and  evaluating  at 280
    %and 360.
If ranges are changed, the .m-file would have to be edited to
change the affected equations.  As an alternative, a general form
of the .m-file could be easily written with the ranges specified
                               269

-------
as a 7 x 2 input matrix.  Note, however, the ranges would have to
be input each time the transformation were invoked.

2.1 Centered Fuel Variables

The transformation outlined above (invtran4.m)  transforms the
random numbers in the interval [-0.5, 0.5]  into uncentered or raw
forms of the fuel variables.  Later, the transformed values will
be used as inputs to model equations.  Generally, the equations
in these equations apply to centered variables.  That being the
case, it is necessary to center the outputs from invtran4.m
before using those outputs as inputs to candidate model
equations.  This, of course, can be easily accomplished simply by
subtracting the center value for a given fuel parameter from all
2000 random values in the column representing that variable.

In Matlab, of course, this is a subtraction of a matrix of center
values from a matrix of random numbers.  The .m-file centerit.m
performs this task,  and is listed below.  The "ones" and "kron"
operation simply put the center values into a conformable matrix
having the same dimensions as the matrix of random "raw" values
of the fuel variables.

                         Table IV-A-4
 CENTERIT.M
 function y=centerit(x,c)
 % where x is a matrix of RAW  (UNCENTERED) fuel parameters
 % c is a column vector of centering values
 % and y is the corresponding matrix of CENTERED fuel
 % parameters
  [q,junk]=size(x);
 wons=ones(q,1);
 z=kron(c',wons);
 y=x-z;
 %This .m-file centers raw variables by subtracting
 %centering values.	
2.2 Uncentered Fuel Variables

There is no reason that the model equations have to be in
centered form.  If the model coefficients apply to uncentered
fuel variables, then the centering transformation  (by means of
centerit.m) would not be used.
3.  COMPUTING EMISSIONS FOR THE RANDOM BLENDS

It is assumed that separate model equations have been developed
for each of the tech groups/emission classes.  These equations

                               270

-------
consist of a set of coefficients for 35 algebraic terms
consisting of 7 linear terms, 7 quadratic terms and 21
first-order interaction terms.  Inasmuch as many of these terms
may not even have been included in the model, the excluded terms
will have zero coefficients.  This fact in no way invalidates
what is to follow because the zero coefficients will simply
eliminate any contribution from the excluded terms.

It is important to note that the invtran4.m and centerit.m
transformations produce only the linear-term values for the
random blends.  It is necessary, therefore, to generate the
quadratic columns of the moment matrix by squaring all entries in
the linear-term columns and annexing these 7 columns to the 7
columns representing the linear terms.  Likewise, it is necessary
to compute all cross-products of the linear columns and to annex
these 21 columns to the 14 columns accumulated at this point.
Note that this procedure is exactly the same as for the
untransformed random numbers in the interval [-0.5, 0.5].   Note,
also, that if raw or uncentered variables are used, a similar
operation would apply to the raw form of the fuel variables.

What remains is to evaluate each of the group equations for the
2000 random blends by substituting the fuel-parameter values for
each blend into the model equation for that group.  In Matlab,
this is conveniently performed as a matrix multiplication, as
shown in the .m-file simple4.m listed below.  Note that this
routine requires as inputs only the matrix of blends to be
evaluated and the matrix of model coefficients.  This coefficient
matrix consists of 35 rows  (one for each term in the equation)
and as many columns as there are tech groups and emission
classes.  The routine has, built-in, the annexation of the
required quadratic and interaction terms as well as an
exponentiation step required to transform the results from log
space to grams/mile space.

                           Table IV-A-5
     SIMPLE4.M
     function y=simple4(x,c)
     %where x is a matrix of the blends to be evaluated
     %with x in raw, centered or standardized form
     %and c is the matrix of the model coefficients
     %with c consistent with the form of x
     %Note: order of variables must be consistent
     %with arrangement of coefficients in c
     %For the EPA iteration4 full model, order of terms is
     % oxygen sulfur RVP E200 E300 aromatics olefins
     x=(x x.rx intergen(x));
     y=x*c;
     Y=exp(y);
     %Note: This is for models having no intercept given.


                               271

-------
As employed in the past, the outputs from the above operation
were weighted and combined into a single set of 2000 quantities
considered as representing the final model output for the
unconsolidated model.  Revision has to be made, however, to
express the model outputs in terms of percent change from a
reference or baseline fuel.  For this purpose, there should be
one more blend in the array input as x.  That blend should be the
reference or baseline blend to which the outputs of the random
blend are to be compared.  It should be expressed in centered or
uncentered form in agreement with the random blends and should be
affixed as the last blend in the list.  The simple4.m algorithm
will perform all necessary computations for this blend, just as
it does for the 2000 or so random blends.

To compute the percent change from baseline, one needs to divide
the unsealed grams/mile for every number in a column by the last
number in that column.  That gives the ratio of the emissions for
the target fuel to the emissions for the reference fuel.  These
numbers then need to be subtracted from 1.0 and multiplied by 100
to yield a percent change from the reference fuel.

The operations described can be executed in Matlab as follows.

First, using the 2001 blends (random 2000 augmented by the
reference blend),  execute the .m-file below.

                           Table  IV-A-6
                    CHANGEI.M
                    function y = change(x)
                    (r,c]=size(x);
                    q=kron(ones(r,1),x(r,:
                    Y=(x./q ) ;	
This .m-file transforms all computed  emissions  into
corresponding ratios to the emission for the reference blend as
evaluated for each vehicle group.   For the subsequent
computation of term contributions to the sum of squares, it is
not appropriate to express the output as percent change.  Many of
the percent changes will be negative, and if one attempts to fit
a regression equation to such data,  one will encounter logarithms
of negative numbers and the output will be in complex numbers.
Rather, one must compute the percent changes from the final
reqression equation in the same way that they are computed for
the vehicle group equations in the unconsolidated model.
4.   COMPUTING TERM CONTRIBUTIONS TO THE MODEL SUM OF SQUARES
                               272

-------
Completion of the steps above yields emissions relative to the
baseline fuel for the 2000 random blends for each of the
technology groups or emission classes.  The baseline fuel is also
included at this point, but should be deleted from the array for
what follows.

We need to fit the 2000 outputs for each of the vehicle groups to
a model consisting of the quasi-orthogonal fuel variables, as
represented in the matrix ucentall.   This matrix consists of 35
columns and 2000 rows, the columns representing the union of the
linear, quadratic and interaction terms.  For the purpose of
examining the relative contributions of terms, we need to square
each of the regression coefficients and divide by an appropriate
normalizing factor.  This normalizing factor is the corresponding
diagonal term obtained from the matrix product ucentall*ucentall.
Some further computations are required to express the term
contributions in percent of cumulative percent, and, of course,
some bookkeeping and display niceties have to be performed.

The basic operation is that of multiple regression.   Any
suitable regression program would suffice, but in the Matlab
environment it is convenient to use a .m-file called reggaer.m.
(See Appendix IV-A.l)

However, as written, this .m-file would compute the regression
coefficients for only one column of input values.  Consequently,
another .m-file was written to cycle the computation through all
columns (i.e., all vehicle groups).   At the same time, the
program combines the group outputs into a composite output and
computes the regression equation for this composite also.  This
.m-file is called sumvoc.m and is listed below.

                          Table  IV-A-7
    SUMVOC.M
    function  [A,C,LIST,SPART]=sumvoc(x,y,wdiag,list,wts)
    y= [y y*wts' ] ,-
         [qr,qc]=size(y);
            LIST=list;
    for i=l:qc,
          [a,c,list,spart]=doitall(x,log(y(:,i)),wdiag,list);
           A=[A;a] ;
           C=[C C] ;
           SPART=[SPART  spart];
    end
The routine incorporates another .m-file called doitall.m, which,
in turn, incorporates the regression program reggaer.m.  The file
called doitall.m is listed below.
                               273

-------
                           Table  IV-A-8
        DOITALL.M
        function [a,c,list,spart]=doitall(x,y,wdiag,list]
        [tpart,apart,f,rr]=reggaer(x,y);
        t=tpart(2:36,1);
        wdiag=wdiag(2:36,:);
        anov=t.*t.*wdiag;
        anovper=100*anov/sum(anov);
        [h,k]=sort(anovper);
        h=h(35:1:1);
        k=k(35:-l:l);
        cumh=cumsum(h);
        a=[sum(anov)  apart (2,1)  rr] ;
        c=[t anov anovper];
        list=list(k,:);
        spart=[k h  cumsum(h)] ;	
First, let us explain the outputs of some of the applicable
.m-files.

The .m-flie reggaer.m can be thought of as an acronym for:

REGression Good And Easy with R-square

Its outputs are:

t-part
a-part
rr
A listing of the regression coefficients, their
standard errors and their t-values

An analysis of variance showing model, residual and
total sums of squares, degrees of freedom and mean
squares

The f-ratio for the model

The r-square for the model
The routine doitall.m augments reggaer.m by incorporating the
operation of squaring and normalizing regression coefficients, as
required to determine the relative contributions of terms.  Note
that,  for this purpose, it requires as input the diagonal
elements of ucentall*ucentall.  Also required as input is an
alphanumeric listing of the terms.  This is the input called
list.

The routine called sumvoc.m simply calls doitall.m repeatedly,
once for each vehicle group,  includinq the composite group.  How
does the composite get in there? As a result of the first line of
code:   y =  [y y*wts].   This instruction says "Annex to the
                               274

-------
columns for the vehicle groups one more, which is just the
weighted combination of the ones you already have." To execute
this instruction, of course, it is necessary to input the
weights, as indicated in the .m-file calling sequence.

Note that all that is required is to invoke sumvoc.m.  It is not
necessary to be concerned with the inner details of what has been
discussed above.  It is important, however, to have all the
inputs:

x         This is a matrix of 2000 rows and 35 columns.  The rows
          are the blends; the columns are the linear, quadratic
          and interaction terms.  For the present application, it
          is simply the matrix ucentall as developed at the
          outset of this manual

y         This is the set of emissions or relative emissions for
          the vehicle groups, as computed by simple4.m or as
          modified by change.m.

wdiag     This is a column vector consisting of the normalizing
          factors obtained by multiplying ucentall*ucentall

list      This is a string variable giving a list of the terms
          represented by the 35 columns of the x-matrix.

wts       This is a row vector of the weighting factors for the
          vehicle groups.

The program sumvoc.m is, among other things, a summarizing
program.  Its outputs,  as specified in the code, are: A, C, LIST
and SPART.  These are explained below:

A         Consists of three columns and as many rows as there are
          vehicle groups.  The first column is the sum of the
          cont n'butions of the term sums of squares for each of
          groups.  The second column is the total sum of squares
          as computed by the ANOVA part of reggaer.m.  The third
          column is the correlation coefficient for the vehicle
          group.  It will be ideally 1 for all but the last row,
          which represents the composite model and is an
          approximation.  Usually it will be well above 0.99.

C         This matrix consists of 35 rows and three times as many
          columns as there are vehicle groups.  The first column
          of each set of three consists of the regression
          coefficients for the corresponding vehicle group.  The
          second column of each set of three consists of the
          actual contributions to the sum of squares for the
          various terms.  The third column of each set of three
          expresses these contributions as percent of the total
          sum of squares.

                               275

-------
LIST      LIST is just that: a listing of the model terms in the
          order in which they were entered in the model.

SPART     This, like C, is a matrix of 35 rows and three times as
          many columns as there are vehicle groups.  This output
          summarizes the results of sorting contributions to sums
          of squares in decreasing magnitude and gives a
          cumulative view.  The first column is simply a listing
          of the term numbers as ordered according to decreasing
          contribution to the sum of squares.  The second column
          gives the percent contributions of the ordered terms,
          and the third column gives the cumulative percent
          totals.

At this point all the quantities of interest have been computed.
However, the display and presentation of results leaves something
to be desired.  Therefore, a so-called "script" .m-file was
written for this purpose; it is called compsumB.m  (see Appendix
IV-A.2).  Note that it requires no inputs; one simply types
compsumB and presses Enter.  The output is a formatted display of
results in a form which permits easy interpretation.

For compsumB.m to operate, however, the results of the previous
computation from sumvoc.m must be available.  Also, those outputs
must be specifically labeled: a, b, c, d.  Therefore, if
compsumB.m is going to be used, one must call for sumvoc.m as
follows: [a,b,c,d]= sumvoc( $,$$,$$$,$$$$,$$$$$),  where $,  $$,
$$$, $$$$,  and $$$$$ are the names that have been given to x, y,
wdiag,  list and wts, respectively.  In other words, the input
quantities can be given any names appropriate to identify their
origin, but the output quantities must be labeled a, b, c and d
if compsumB.m is going to be used for display.
5.   COMPUTING THE FINAL MODEL EQUATION

After the term contributions have been reviewed and a decision
has been made as to what terms are to be retained, the original
data are refitted to an equation of the selected form and a set
of revised vehicle-group equations are returned.

The steps above are usually repeated, but the main objective is
the final, composite equation.  However, this equation needs to
be put in terms of either centered or uncentered coefficients,
not coefficients for the quasi-orthogonal terms used to compute
term contributions.

Centered coefficients can be computed simply by fitting the
final, composite relative emissions to the desired terms.  Any
suitable regression program can be used.
                               276

-------
Using reggaer.m, one can compute the centered coefficients by the
following code:

                           Table  IV-A-9
                    [tcent,acent,fcent,rcent]=
          reggaer([xcent xcent.*xcent intergen(xcent)],
          	log(yvoc*wtsvtox') )	
where yvoc is the final set of computed emissions for the
composite, unconsolidated model and wtsvtox is a row matrix of
weighting factors for the vehicle groups.

If one desires the uncentered coefficients, simply change the
above code to:

                          Table IV-A-10
                     [traw,,arawt,fraw,rraw]=
            reggaer([xraw xraw.*xraw intergen(xraw)],
            	log(yvoc*wtsvtox') )	
Percent change for a target blend can then be computed by
computing the yield for the target and for the baseline blend
using the equation:
6.
 100*{[(Target yield)/(Baseline yield)]  - 1}.

SOME CAVEATS
The procedures outlined above are tailored to a data set in which
there are seven fuel variables and a total of 35 terms to be
considered.  If fewer terms are to be included in the model, the
extraneous terms can be assigned zero values.  However, in the
case of toxics, no interaction terms are involved; consequently,
it may be cumbersome to carry along extra columns in the matrix.
A simpler version of the appropriate .m-files can be used with
suitable modification.  On the other hand, some of the toxics
have more than seven variables (such as separate terms for MTBE,
Ethanol, ETBE and TAME).   In these cases, it will be necessary to
start with a larger matrix of random input values.

It is, of course,  possible to put all of the above computations
into a very general form which would accomodate all cases, but
that has not been done at this point.  Such generalization would
have both advantages and disadvantages.  It would make for a
unified approach,  but it might necessitate carrying along a large
number of irrelevant quantities.   Suggestions on this point are
solicited.
                               277

-------
H.  T.  McAdams
Advanced Computing Center of Argenta
Argenta, Illinois
8-31-93
                               278

-------
Appendix IV-A.I:  GENERAL REGRESSION ROUTINE


%This .m-file computes the regression equation - i.e., intercept
%and slopes, and computes t for testing the significance of the
%slopes of the regression line against the null hypothesis
%that each slope = 0.
%It also provides an analysis of variance, an f-ratio and r
square.
%The ANOVA gives sum of squares, degrees of freedom
%and mean squares for the mean, regression and residuals.
function [tpart,apart,f,rr]=reggaer(x,y)
     [u,v] =size(x) ;
     x=[ones(u,1) x] ;
     tt=x'*x;
     q=inv(tt);
     b=q*x'*y;
%b is the vector of regression coefficients: intercept and slope
      z=y-x*b;
        [h,k]=size(b);
     tl=z'*z;
%tl is the residual sum of squares
     t2=u-h;
%t2 is the residual degrees of freedom
     t3=tl/t2;
%t3 is the residual mean squares
      z=z/(u-h);
%z is the error variance for the regression line
      varb=t3*q;
%varb is the variance of the regression coefficients.
stdb=sqrt(diag(varb));
t=abs(b./stdb);
sl=y'*y;
%sl is the totel sum of squares
[q,r] =size(y) ;
    s2=q;
%s2 is the total number of degrees of freedom
    s3=sl/s2;
%s3 is the total mean squares
      ql=sum(y)*sum(y)/q;
%ql is the sum of squares for the mean
     q2 = r;
%q2 us the degrees of freedom for the mean
     q3=ql/q2;
%q3 is the mean squares for the mean
    wl=b'*x'*y-ql;
%wl is the regression sum of squares
    w2=h-q2;
%w2 is the regression degrees of freedom
    w3=wl/w2;
%w3 is the regression mean squares
      f=w3/t3;

                               279

-------
sumsq=[ql wl tl si]1;
df=[q2 w2 t2 s2] '  ;
meansq=[q3 w3 t3  s3]';
b=b' ;
stdb=stdb';
t=t';
sumsq=sumsq';
df=df';
meansq=meansq' ;
tpart=[b1 stdb' t' ] ;
apart=[sumsq1 df'  meansq']
f = f;
rr=wl/(wl+tl);
                                280

-------
Appendix IV-A.2:  SCRIPT FILE FOR DISPLAYING RANDOM BALANCE
OUTPUT
Note: This display is constructed for 10 vehicle groups.  As
used, the 8th group is a dummy.
COMPSUMS . M

sp=
al='The following compares the model sum of  squares  ';
a2='as computed as a weighted sum of coefficients  squared';
a3=' (first column) with the model sum of squares';
a4= ' computed from regular ANOVA  (second column) ';
aa='Third column is R-square';
disp (al)
disp(a2)
disp (a3)
disp (a4)
disp (aa)
disp (sp)
a(:,l:3)
aS='The following displays, for each of the  groups';
a6='the regression coefficients  (column 1) ' ;
a7='the termwise contributions to model sum  of  squares  (column
2) ';
a8='the percent termwise cont . to model sum  of  squares  (column
3) ';
a9=' Order of terms is as follows:';
disp (sp)
disp (a5)
disp (a6)
disp(aV)
disp (a8)
disp(a9)
disp (sp)
disp (c)
disp (sp)
gl= 'Group 1 ' ;
disp(gl)
g2= 'Group 2 ' ;
disp(g2)
b(:,4:6)

g3= 'Group 3 ' ;
disp(g3)
b(:,7:9)

g4= 'Group 4 ' ;

                               281

-------
disp(94)
b( : ,10:12)

g5= 'Group 5 ' ;
disp(g5)
b(:, 13:15)

g6 = 'Group 6 ' ;
disp (g6)
b( : ,16:18)

g7= 'Group 7 ' ;
disp(gV)
b( : ,19:21)

g8= 'Group 8 ' ;
disp (g8)
b( : ,22:24)

g9= 'Group 9 ' ;
disp (g9)
b( : ,25:27)

glO= 'Group 10 ' ;
disp(glO)
b( : ,28:30)

gll= 'Weighted Composite';
disp(gll)
b(:, 31:33)

w3='The following gives, for each group,  the  term number';
w4= ' ranked in decreasing order of importance  (column 1) ';
w5='the percent contribution to model  sum of  squares (column 2) ';
w6='and the cumulative percent contribution to
     model SS (column 3) ' ;
w7= ' followed by an ordered listing of  the term  names';
disp (w3)
disp (w4)
disp (w5)
disp (w6)
disp (w7)

disp (sp)
alO=c;
hl= 'Group 1 ' ;
disp (hi)
h2 = ' Group 2 ' ;
disp(h2)
                               282

-------
d(:,4:6)
alO(d(:,4) , :)

h3 ='Group  3';
disp(h3)
d(:,7:9)
alO(d(:,7) , :)

h4 ='Group  4';
disp(h4)
d(:,10:12)
alO(d(:,10) , :)

h5='Group  S';
disp(h5)
d(:,13:15)
alO(d(:,13) , :)

h6='Group  6';
disp(h6)
d(:,16:18)
alO(d(:,16) , :)

h7='Group  7';
disp(h7)
d(:,19:21)
alO(d(:,19) , :)

h8='Group  8';
disp(h8)
d(:,22:24)
alO(d(:,22) , :)

h9='Group  9';
disp(h9)
d(:,23:27)
alO(d(:,2S) , :)

hlO='Group 10' ;
disp(hlO)
d(:,28:30)
alO(d(:,28) , :)

hll='Weighted  Composite';
disp(hll)
d(:,31:33)
alO(d(:,31) , :)
                                283

-------
Appendix IV-B:  Glossary and Nomenclature
BZV%
DVOC

EtOH
GM
HDGV
HDV
HSVOC

I&M
LDGT
LDGV
LDT
LDV
MTBE
POM
RFC
RFVOC

RLVOC

OXY
OLE
SUL

ARO/AROM
E200
E300
BEN
RVP
VMT
VOC
A,B
At,Bt
Ab,Bb
Aet

Bet

C
E(t)
          Volume percent of benzene in the fuel
          Total grams per mile of volatile organic compounds from
          diurnal emissions
          Ethanol
          General Motors
          Heavy-duty gasoline vehicle
          Heavy-duty vehicle
          Total grams per mile of volatile organic compounds from
          hot soak emissions
          Inspection and maintenance; also IM, I/M
          Light-duty gasoline truck
          Light-duty gasoline vehicle; see LDV
          Light -duty truck
          Light-duty vehicle, equivalent to light-duty car
          Methyl tertiary-butyl ether
          Polycyclic organic matter
          Reformulated gasoline
          Total grams per mile of volatile organic compounds from
          refueling emissions
          Total grams per mile of volatile organic compounds from
          running loss emissions
          Total oxygen content of fuel in terms of weight percent
          Olefins content of fuel in terms of volume percent
          Sulfur content of fuel in terms of parts per million by
          weight
          Aromatics content of fuel in terms of volume percent
          Percent fuel distilled at 200°F
          Percent fuel distilled at 300°F
          Fuel benzene content in terms of volume percent
          Reid vapor pressure in terms of psi
          Vehicle miles travelled
          Volatile organic compounds, measured in grams per mile
          on a nonme thane, none thane basis

          =     Generic fuel parameters
          =     Generic fuel parameters  for a target fuel
          =     Generic fuel parameters  for a baseline fuel
          =     Value of  generic fuel parameter A at the end of
                the allowable range
          =     Value of  generic fuel parameter B at the end of
                the allowable range
          =     Average emissions of  any pollutant from vehicle
                group i
          =     Unknown gram/mile emissions for vehicle group  i,
                resulting from statistical regressions that
                disregard intercept  terms
          =     Unknown gram/mile emissions for vehicle group  i
                for a specific target fuel
                               284

-------
EL (b)      =     Unknown gram/mile emissions  for vehicle group  i
                for a specific baseline  fuel
f.±         =     Sum of fuel terms * coefficients  for vehicle
                group i for any  fuel
f^t)      =     Sum of fuel terms * coefficients  for vehicle
                group i where the fuel terms  represent a
                specific target  fuel
f± (b)      =     Sum of fuel terms * coefficients  for vehicle
                group i where the fuel terms  represent a
                specific baseline fuel
f(t)       =     Second or  first  order function of  fuel
                parameters for a target  fuel
f (b)       =     Second or  first  order function of  fuel
                parameters for a baseline  fuel
ki         =     Proposed adjustment factor for vehicle group i
PJ         =     Any Complex Model fuel parameter
VS±         =     Proposed weighting factor  for vehicle group i
v!±         =     Final weighting  factor for vehicle group  i
X         =     Fleet average grams/mile of  any pollutant for
                any fuel
X(t)       =     Fleet average grams/mile of  any pollutant for  a
                specific target  fuel
X.±(t)      =     Grams/mile of any pollutant  for a  specific
                target fuel for  vehicle  group i
X(b)       =     Fleet average grams/mile of  any pollutant for  a
                specific baseline fuel
X^b)      =     Grams/mile of any pollutant  for a  specific
                baseline fuel for vehicle  group i
Y         =     Fleet average performance  of  any pollutant for
                any fuel
Y(t)       =     Fleet average performance  of  any pollutant for a
                specific target  fuel
Y^t)      =     Performance of any pollutant  for a specific
                target fuel for  vehicle  group i
Y(b)       =     Fleet average performance  of  any pollutant for a
                specific baseline fuel
Y^b)      =     Performance of any pollutant  for a specific
                baseline fuel for vehicle  group i
Y         =     Emissions  of any exhaust pollutant in percent
                change from a baseline fuel
Y(t)       =     Emissions  of a generic exhaust pollutant  from  a
                target fuel
Y(b)       =     Emissions  of a generic exhaust pollutant  from  a
                baseline fuel
ek         =     Coefficient determined through statistical
                regressions
*         =     Denotes a  multiplication symbol.   For example,
                A*B means  the quantity A multiplied by the
                quantity B
                               285

-------
Appendix IV-C:  MOBILE4.1 Input File Used to Develop the Simple
                Model and the Phase I Complex Model  (must be run
                with interim version of MOBILE4.1)

1          PROMPT -
MOBILE4.1 v.7/11/91: Simple Model correlations
1          TAMFLG -
1          SPDFLG -
1          VMFLAG -
1          MYMRFG -
1          NEWFLG -
2          IMFLAG - enter I/M program
1          ALHFLG -
2          ATPFLG - enter ATP
2          RLFLAG - Stage II refueling controls
2          LOCFLG - enter LAP record once
1          TEMFLG -
3          OUTFMT - print 112 column descriptive output format
4          PRTFLG - print exhaust HC,  CO and NOx results
1          IDLFLG -
3          NMHFLG - print VOC
3          HCFLAG - print HC components
83 20 68 20 00 00 100 1 1 2221 2 11 20 20 20
83 68 20 2221 11 100. 11111212
89 1 086 086
RegNeg           C 71.6 91.6 11.5 08.7 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 08.5 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 08.3 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 08.1 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.9 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.7 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.5 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.2 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           C 71.6 91.6 11.5 07.0 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 08.7 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 08.5 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.8 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.6 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.4 89 1

                               286

-------
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.2 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 07.0 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 06.8 89 1
1 15 19.6 75.0 20.6 27.3 20.6
RegNeg           B 69.4 94.0 11.5 06.6 89 1
1 15 19.6 75.0 20.6 27.3 20.6
                               287

-------
Appendix IV-D:  MOBILE5a Input File Used To Develop the Phase II
                Complex Model

I          PROMPT -
Phase II Complex Model baseline runs for LDGVs & LDGTs
1          TAMFLG -
1          SPDFLG -
1          VMFLAG -
1          MYMRFG -
5          NEWFLG - no CAA mandates are modeled
2          IMFLAG - enter I&M program
1          ALHFLG -
8          ATPFLG - enter ATP; include pressure and purge checks
2          RLFLAG - enter Stage II
1          LOCFLG -
1          TEMFLG -
3          OUTFMT - print 112 column descriptive output format
1          PRTFLG - print exhaust & evap HC only
1          IDLFLG -
3          NMHFLG - print VOC
2          HCFLAG - no HC components
83 20 68 20 03 03 096 1 1 2221 4211 0.80 20.0 2.00 Enhanced IM
83 84 20 2221 11 096. 12211111                     ATP
83 83 20 2221 11 096.                              Pressure test
83 86 20 2221 11 096.                              Purge test
89 1 80. 60.                                       Stage II
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 09.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 09.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 09.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 09.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 09.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 08.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	 C 71.6 91.6 09.5 07.5 90 1 1 1

                               288

-------
1 15 19.6 87.0 20.6 27.3 20.6  1
	  C 71.6 91.6 09.5 07.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  C 71.6 91.6 09.5 07.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  C 71.6 91.6 09.5 06.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  C 71.6 91.6 09.5 06.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  C 71.6 91.6 09.5 06.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 09.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 09.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 09.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 09.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 09.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 08.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.5 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.3 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 07.1 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 06.9 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 06.7 90 1 1 1
1 15 19.6 87.0 20.6 27.3 20.6  1
	  B 69.4 94.0 09.5 06.5 90 1 1 1
                               289

-------
Appendix IV-E:  SAS code used to Devlop Raw Exhaust Emission
Models
                               290

-------
Appendix IV-F:  Included 1990 Technology Vehicles in the Complex
                Model Database

     The list below shows the 1990 technology vehicles that are
currently in the complex model database used to develop the
exhaust models shown in Section A.
                               291

-------
V.   Environmental and Economic Impact of Phase I Reformulated
     Gasoline

     The Phase I reformulated gasoline standards consist of two
fuel specifications (maximum benzene content and minimum oxygen
content) and three performance standards applying to VOC,  toxics
and NOx emissions reductions required of reformulated fuels from
1995-1999.  After 1999,  the more stringent Phase II emission
performance standards, discussed in Section VI,  take effect.
EPA's April 1992 reformulated gasoline proposal extensively
discussed the development of the Phase I simple model standards.
Phase I complex model standards and other Phase I issues were
discussed in the February 1993 proposal.  Comments on and any
further developments of the Phase I standards,  either under the
Simple or Complex Models, were discussed in Sections 2 and 4
above.  This section presents both the costs and benefits of the
Phase I standards.

A.   Introduction

1.   Phase I Fuel Control Costs

     The cost to produce Phase I RFG was determined for Regions
1, 2 and 3 through the recent refining cost analysis done by
Bonner and Moore (B & M) .51  Unlike the Phase II data, individual
costs for each parameter change (added oxygenate, reduced benzene
and reduced RVP) were not determined in the B & M cost analysis.
The cost of the phase I RFG program was calculated in one step
for all three parameter changes.  Because the RFG specifications
differ between summer and winter,  the B & M refinery model runs
separate summer and winter optimizations off the same base of
capital equipment.  For accounting purposes, the model allocates
all the capital costs only to the winter.  This produces
appropriate and accurate estimates of annual average costs,  but
makes it difficult to develop separate winter and summer costs.
     51Bonner and Moore  Management  Science,  "Study of the
Effects of Fuel Parameter Changes  on  the  Cost  of  Producing
Reformulated Gasoline," Prepared for  EPA  under contract  through
Southwest Research Institute and the  National  Institute  for
Petroleum and Energy Research.  This  data was  made available to
the public through the  following document:   "DOE  and API Phase
II Cost Estimates," EPA Memorandum from Lester Wyborny,  FSSB,
to the Air Docket, November 4,  1993.   The Bonner  and Moore work
performed for this analysis was based on  a  revised version of
the model used to generate costs for  the  draft RIA.   This new
version includes provisions to  prevent dumping of fuel
components into conventional gasoline.

                               292

-------
Therefore, the winter and summer costs are shown averaged
together here and represent an annual average per gallon cost.

     The costs developed below exclude the cost of oxygenates
already required to be used in winter oxyfuel areas.  This was
done because the winter oxyfuel program began prior to the RFC
program and because this oxygenate produces substantial CO
emissions reductions which justify the cost of the oxygenate.

     The first column in Table V-l below summarizes the
manufacturing costs estimated by the refinery model of producing
Phase I RFC.  In the adjacent columns of that table, the capital
portion of these costs were then adjusted to reflect a 10 percent
rate of return on capital from a 15 percent basis.  In Table V-2
further below, the costs were further adjusted to account for the
change in fuel's heat content; the fuel economy effects were due
to the added oxygenates and reduced RVP.  The effect of fuel
composition on vehicle fuel economy is addressed in the analysis
of the Phase II standards in Section VI.

          Table  V-l:  Phase  I  RFC  Costs  Based  on  10%  ROI
Region
Region 1
Region 2
Region 3
Unadjusted
Phase I
Costs (15%
ROI) (


-------
     Given the mandate for at least 2.0 volume percent oxygen,
the cost of producing Phase I RFC is partially determined by the
cost of adding oxygenates.  The cost of oxygenate addition,
however, is largely driven by the purchase cost of the oxygenate,
or the raw materials used in producing the oxygenate.  When the B
& M model was being set up to estimate the RFC program costs,  the
oxygenate costs were programmed into the model based on the
expectations for the future market of oxygenate demand and supply
at that time.  However, recent indications are that the oxygenate
costs may vary significantly from these earlier estimates and
that the cost of producing Phase I RFC should reflect these
varying oxygenate costs.  Furthermore, much of the ethanol that
the model estimated would be blended into RFC was done so at an
unsubsidized cost (retail cost without the 54 c/gal subsidy).   An
unsubsidized ethanol cost was used because, at that time, ethanol
blends under 10 percent ethanol were not permitted to take
advantage of the subsidy.  Since then, the subsidy policy has
been modified and ethanol blends at less than 9 volume percent
are eligible for the excise tax credit.  EPA is therefore
adjusting these costs to reflect the tax credit available to all
ethanol blends.

     All of the oxygenate cost projections were made by B & M in
a period of strong growth in oxygenate demand and when the
possibilities of supply just meeting demand were real.  Thus,  the
B & M projections tend to be quite high in today's terms.  For
example, B & M projected a merchant MTBE price of $1.05/gal.
Despite this being the beginning of the 1993-94 winter oxyfuel
season, MTBE has been selling for a few months on the spot market
for $0.66-0.75/gal.   The summer runup in prices which occurred
during 1992 while refiners built up inventories for the winter
never materialized in 1993.  The same could happen again with the
onset of the reformulated gasoline program.  Oxygenate prices
could be relatively high in late 1994 and early 1995 and then
fall back to more reasonable levels.  To account for this, a
range of MTBE and methanol prices is being used, as shown in the
table below.   The low end of the ranges reflect today's low
energy prices and ample oxygenate supply.  The high end of the
ranges are those projected by B & M.

     For ethanol, B&M projected the historical figure of
$1.20/gal, prior to consideration of the tax credit.  This price
allows ethanol to be competitive as a gasoline extender and
doesn't reflect its octane or oxygenate value.  During the summer
of 1992, ethanol prices reached $1.40/gal and higher.  Recently,
with wholesale regular gasoline prices in the mid-40's per
gallon, ethanol has been selling for $1.00-1.05/gal.  To be
consistent with the MTBE projections, the range of $1.20-1.60/gal
will be used here, or $0.66-1.06/gal after factoring in the
$0.54/gal excise tax credit.  These are plant gate prices and
generally can be used as is throughout the Midwest.  However,  use
in reformulated gasoline, either directly or in producing ETBE,

                               294

-------
outside the Midwest would involve a significant shipping cost.
To account for this, distribution costs for shipping to the
Northeast and Gulf coast of 5^/gal were added onto the price of
ethanol, respectively.

     Regarding other oxygenates, TAME was not purchased in the
refinery modeling, but was produced within the refinery.  Thus,
the range of methanol prices is adequate to describe its
potential cost to refiners.  Also, neither ETBE production nor
purchase was considered given the technological and economic
uncertainties existing at the time of the study.  In order to be
utilized in significant quantities, ETBE would have to be
competitive with MTBE and ethanol, and would thus yield roughly
equal or lower costs than those determined below.  Table V-3
below summarizes the oxygenate prices used in the model and the
oxygenate price range established by EPA.
     Table V-3: Oxygenate Costs used in the Phase I Analysis
Oxygenate
Ethanol
MTBE
Methanol
Model Costs
$1.20
$1.05
$0.57
Cost Range
Before Subsidy $1.20
After Subsidy $0.66 -
- $1.60
$1.06
$0.80 - $1.05
$0.40 - $0.57
     The Phase I costs for Regions 1, 2 and 3 shown in Table V-2
were then adjusted to reflect the range of possible oxygenate
costs.  The results are summarized in Table V-4.  The regional
Phase I cost estimates were combined into a U.S. cost by
weighting each regional cost by the fraction of RFG produced by
that region (see Table VI-# in Section VI) .
                Table V-4:  Final  Phase  I  RFG  Costs
Geographic Area
Region 1
Region 2
Region 3
U. S.
Cost Range
(c/qal)
2.6 - 4.1
2.3 - 6.5
3.5 - 4.6
3.0 - 4.9 (3.9
average)
                               295

-------
     Total annual cost to the nation would simply be the cost per
gallon from Table V-4 multiplied by the fuel consumption in the
reformulated gasoline areas plus any spillover into adjacent
areas.  The Energy Information Agency projects total national
motor gasoline consumption to be 113.9 and 119.8 billion gallons
per year in 1995 and 2000, respectively.  Interpolating, the
estimate for 1998 would be 117.5 billion gallons per year.  In
the draft RIA for the February NPRM, reformulated gasoline was
estimated to represent 32.1 percent of national fuel consumption,
including a 10% spillover rate.  (Of this total, 13.3% was in
Class B areas and 18.8% was in Class C areas.)  Combining these
figures yields a total annual cost of $1.13-1.85 billion.
2 .
Emissions Reductions
     In-use emission reductions for calendar year 1998 were
determined using both MOBILES.0 and the complex model.  Emission
reductions were estimated under both basic and enhanced I/M
scenarios because both of these I/M programs will be in place in
the 1995-1999 period.  MOBILES.0 was used to estimate baseline
VOC emissions and the effect of RVP on non-exhaust VOC emissions,
as was done in the Draft RIA.  The complex model was used to
estimate the effect of fuel composition on exhaust VOC emissions.
Derivation of the in-use baseline VOC emission projections is
described in more detail in Section III.B of the Draft RIA and
the baselines themselves are shown in Tables A1-A4 of that
section of the DRIA.

       Table  V-5:  1998  In-Use Baseline  VOC  Emissions  (g/mi)

Exhaust VOC
Nonexhaust VOC
Total VOC
Basic I/M Program
Region 1
Region 2
1.456
1.449
0.816
1.006
2 .272
2.455
Enhanced I/M Program
Region 1
Region 2
0.991
0.991
0.570
0.681
1.561
1.672
     Assuming fuels meeting the averaging requirements for Phase
I RFG under the Simple Model,  and the effects of oxygen addition
and RVP and benzene reduction on other fuel parameters, as
developed in Section VI,  the Phase I fuel compositions shown in
Table V-6 were developed.
                               296

-------
    Table V-6: Baseline and Phase I In-Use Fuel Compositions

Fuel Parameter
Oxygen
Sulfur
RVP (psi)
E200 (vol%)
E300 (vol%)
Aromatic s (vol%)
Olefins (vol%)
Benzene (vol%)
In-Use Baseline
Fuel
0.0
339
7.8/8.7 (Region 1/2)
41.0
83.0
32.0
9.2
1.53
Phase I In-Use Fuels
Region 1
2 .1
309
7.1
46.7
84.9
25.5
13.1
0.95
Region 2
2 .1
309
8.0
46.7
84.9
25.5
13.1
0.95
     The Phase I complex model was used to evaluate the effect of
these fuels on VOC emissions.  The resulting percent reductions
were then applied to the baseline VOC emissions shown above.  The
results are shown in Table V-7.
     Table V-7: 1998 In-Use Phase I RFG VOC Emissions (g/mi]

Exhaust VOC
Nonexhaust VOC
Total VOC
Basic I/M Program
Region 1
Region 2
1.300
1.294
0.617
0.765
1.917
2.059
Enhanced I/M Program
Region 1
Region 2
0.886
0.886
0.443
0.535
1.330
1.421
     The differences in the VOC emissions of Tables V-5 and V-6
represent the benefits of the Phase I RFG program.  No
substantial NOx emission reduction is expected.  To convert the
g/mile emission reductions to annual tons, a 1998 national fuel
consumption of 113.84 billion gallons was assumed, along with an
average gasoline vehicle fuel economy of 20.58, from the
MOBILE4.1 fuel consumption model.  It was also assumed that RFG
would be sold in the 9 mandated areas plus those areas which had
opted in by mid-summer 1993.  The result is that approximately
                               297

-------
18.2 and 12.8 percent of national fuel consumption will be
reformulated gasoline in Class C and B areas,  respectively.

     A significant issue is the calculation of the VOC emission
reduction.  The VOC controls for reformulated gasoline only apply
in the summer, or high ozone season.  Most other VOC control
programs apply yearround, due to the permanent nature of the
controls  (e.g., modified painting facilities,  motor vehicle
control, I/M programs, etc.).  The effectiveness of these
programs is typically reported on an annual VOC control basis;
that is, the emission reductions estimated to occur include VOC
emissions reduced in the winter as well as the summer, even
though the winter VOC emission reductions have little ozone
related value.  In order to be comparable with these programs,
the benefits of RFG must be put on an annual basis.  This means
extending the real benefits occurring in the summer to the rest
of the year.  This is appropriate since the RFG summer VOC
emission benefits provide the same in-use ozone reductions as a
yearround program which reduces summer emissions to the same
degree but also reduces emissions in the winter.  To do this, the
gram per mile emission reductions were applied to yearround fuel
consumption.
        Table V-8:
VOC Emission Reductions of Phase I RFG
   (annualized tons/yr)

Class B
Class C
Total
Basic I/M
118,000
187,000
305,000
Enhanced I/M
77,000
119,000
196,000
     Phase I RFG also reduces cancer incidence due to the
reduction in air toxics emissions required.  Section VI outlines
a methodology for estimating baseline and controlled toxics
emissions from MOBILES and the complex model.  It also outlines a
methodology for converting emissions of air toxics into cancer
incidence.  That methodology was used here.  The analysis of the
Phase II RFG standards in Section VI also presents a cancer
incidence reduction associated with the Phase I standards as a
starting point from which to conduct the Phase II analysis.

     Only three adjustments of the results of the Phase II
analysis need to be made to use the results directly here.  The
first two refer to baseline emissions.  First, the Phase I
analysis focuses on the year 1998, while the Phase II analysis
focuses on 2003.  Second, this Phase I analysis considers the
fact that both basic and enhanced I/M programs will exist prior
to 2000, while only enhanced I/M is considered thereafter.  The
                               298

-------
result is that the toxics emissions benefits estimated in Section
VI, as well as the resulting cancer incidence reduction,  need to
be increased to reflect the difference between the baseline VOC
emissions in Table V-5 above and those estimated in Section VI.

     The third difference is that the Phase II analysis assumes
that the fuel modifications will only occur in the summer.  On
the other hand, the Phase I benefits are primarily due to the
addition of oxygenate  (with its effect on aromatic content) and
the reduction of fuel benzene content and will occur yearround.
(The toxics benefits of the summer RVP control are very slight
and can be ignored in this extrapolation from the Phase II
analysis.)   Taking all this into account, the Phase I RFC
standards will reduce U.S. cancer incidences (assuming one year
exposure to both the baseline and controlled levels of emissions)
by 24 (assuming basic I/M in place) and 16  (assuming enhanced I/M
in place) incidences.
                               299

-------
VI.  Phase II Reformulated Gasoline Standards and NOx Standards
     for Reformulated Gasoline
A.   Introduction

     The development of the VOC and toxics emission reduction
standards for Phase II reformulated gasoline (RFC),  as well as
the NOx emission reduction standards which EPA is also
promulgating, is presented here.  The statutory requirements,
analytical methodologies,  and results, as well as public comments
received on the proposal and EPA's responses, are discussed
below.  The Phase II RFC standards and NOx standards can be found
at §80.41 and are summarized in section VI of the final
rulemaking preamble.

1.   Statutory Requirements

     Section 211(k)  requires that reformulated gasoline achieve
the greatest reductions possible in volatile organic compounds
(VOCs) and toxics emissions, "taking into consideration the cost
of achieving such emission reductions, any nonair-quality and
other air-quality related health and environmental impacts and
energy requirements.  Specifically, section 211(k)(3)(B) of the
Act requires that,  in the year 2000 and beyond, "aggregate
emissions of ozone-forming volatile organic compounds from
baseline vehicles52 when using reformulated gasoline shall be 25
percent below the aggregate emissions of ozone forming volatile
organic compounds from such vehicles when using baseline
gasoline53."  Similarly, a 25% reduction in emissions of toxic
air pollutants is required.  The Act also specifies that the
Administrator may adjust the 25 percent reduction level to
provide for lesser or greater reductions based on technological
feasibility, giving consideration to the cost of achieving such
reductions.  In no case can the required reduction be less than
20 percent.  The Act further provides that emissions of oxides of
nitrogen  (NOx)  cannot increase as a result of the use of
reformulated gasoline.  These VOC and toxics reductions and NOx
limit are known as the Phase II reformulated gasoline standards.

     Section 211(c)  of the Act allows the Administrator to
regulate fuels or fuel additives if "any emission product of such
     "According  to  section 211(k)(10)(A)  of the Act, "baseline
vehicle" means representative  model  year  1990 vehicles.

     53The  formulation for summertime baseline gasoline is
defined in section  211(k)(10)(B)  of  the Act.   See further
discussion of baseline emissions  in  section III.

                               300

-------
fuel or fuel additives causes, or contributes to, air pollution
which may reasonably be anticipated to endanger the public health
or welfare."  Section 211(c)(2)  further provides that EPA cannot
control these fuels and fuel additives "except after
consideration of all relevant medical and scientific evidence
available..., including consideration of other technologically or
economically feasible means of achieving emissions standards."
In addition, EPA must find that the prohibition "will not cause
the use of any other fuel or fuel additive which will produce
emissions which will endanger the public health or welfare to the
same or greater degree than the use of the [regulated fuel/fuel
additive]."

     EPA had elected to use this authority to require
reformulated fuels to also achieve NOx reductions in order to
reduce ozone formation, based on scientific evidence regarding
the benefits of NOx control and on the cost-effectiveness of NOx
reductions.  The determination of the need for and cost-
effectiveness of NOx control is evaluated in this study.  A
discussion of the scientific justification for NOx reductions
under the reformulated gasoline program can be found in section B
below.

2.   Factors to be Considered in Setting the Phase II RFC
     Standards

     An analysis, per the requirements of section 211(k)(1), of
the costs,  cost-effectiveness, health, environmental and energy
impacts of fuel component modifications used to reduce VOC, NOx,
and toxics emissions of baseline gasoline is presented below.
Because the cost of achieving a Phase II reformulated gasoline
standard was specifically mentioned as a consideration in setting
the standard, this analysis focuses on the cost-effectiveness of
different fuel reformulations which achieve at least the minimum
reductions.  Since it has been determined that the health and
environmental impacts of such fuel reformulations are beneficial
and increase as the percentage emission reduction increases, and
that the energy impacts are minimal,  the Phase II reformulated
gasoline performance standards are determined based primarily on
cost-effectiveness.  A discussion of these impacts is presented
in section H below.

     EPA received a comment to the proposed standards which
stated that EPA must determine whether VOC or toxics are the
controlling factor in setting the Phase II RFG standards.   As
will be shown by the following analysis, EPA has determined that
control of toxics below the minimum statutory level  (20%)  is not
cost effective.  Hence, the standards will be set based on the
cost-effectiveness of VOC and NOx control, not on the basis of
toxics control.
                               301

-------
B.    Scientific Justification for NOx Control

     Along with the proposed VOC and toxics standards for Phase
II  reformulated gasoline, EPA proposed NOx standards, based on
the cost-effectiveness and environmental benefits of NOx control.
Many comments were received regarding the NOx standards.  These
comments are summarized below.  Several of the comments
questioned the scientific evidence for the benefits of
controlling NOx to reduce ozone formation.  Hence, a discussion
of  the benefits of NOx reduction in terms of ozone impacts, as
well as other environmental and health benefits, is also
presented below.

     The following discussion summarizes EPA's consideration
under §211(c)(2)(A) of the relevant medical and scientific
evidence available to the agency,  as well as its consideration of
other technologically or economically feasible means of achieving
NOx reductions through vehicle standards under §202.
Furthermore, section H provides a comparison of the cost-
effectiveness of other means of NOx control from mobile or
stationary sources, and concludes that control of NOx emissions
from reformulated gasoline will not pose a greater burden on
industry or the public than these alternative technologies.
Finally, since the complex model allows consideration of all of
the components of gasoline which affect NOx emissions, EPA
believes that the NOx standards (determined using the complex
model) will not lead to the use of any fuels or fuel additives
which will endanger the public health or welfare  (through
increases in VOC,  NOx, or toxics emissions) more than would occur
if  NOx emissions from reformulated gasoline were not regulated.

     EPA has previously established National Ambient Air Quality
Standards for ozone, N02,  and PM-10.   In so doing,  EPA provided
all of the medical and scientific information available to
support the fact that regulation of these pollutants is
beneficial to public health and welfare.  Since NOx emissions
contribute significantly to the inventory of ozone and N02,  and
in some cases,  PM-10, reductions in NOx emissions will lead to
reductions in the inventories of these pollutants.

1.    Comments on Proposed Standards

     EPA received a variety of comments on the proposed NOx
standards.  Some commenters claimed that a NOx reduction standard
was counter to the regulatory negotiation agreement.  The focus
of  Reg-Neg was on how the provisions of section 211(k) were to be
interpreted and implemented; it did not address EPA's authority
under section 211(c) to regulate emissions of NOx.  The
scientific evidence summarized below has convinced EPA of the
need to control mobile source emissions of NOx during the summer
months in ozone nonattainment areas, and EPA is doing so under
the authority granted by section 211(c).  EPA has determined that

                               302

-------
NOx standards for gasoline sold in ozone nonattainment areas
should be coordinated with the Phase II performance standards for
VOC and toxics since the environmental benefits of the combined
programs generally will be as great as  the benefits of the
programs implemented separately,  while costs will be considerably
less.  The special distribution and enforcement requirements will
already be in place for the Phase II standards and thus neither
industry nor government will incur additional costs (beyond
manufacturing costs) as a result of implementing an additional
NOx standard.  Furthermore, EPA has decided to promulgate the NOx
standards with the Phase II RFC standards to reduce the burden on
the industry of complying with two separate regulations within a
short timeframe (which could negatively impact decisions
regarding capital investment if they were not well coordinated).
     Other comments questioned the ability to control NOx
emissions with any parameter other than oxygen.  Since EPA
requires oxygen in reformulated gasoline (as directed by the
Clean Air Act),  one comment suggested that the only way to
control NOx would be to eliminate the requirement of a minimum
oxygen content.   The complex model clearly shows that some
parameters, most notably sulfur,  can have a significant impact on
NOx emissions.  Furthermore, the model indicates that the
addition of oxygen leads to a small (essentially negligible)
decrease in NOx emissions.  While the impact of various
oxygenates on NOx emissions can vary due to other changes in the
fuel which accompany oxygen addition,  the oxygen alone does not
appear to increase NOx in 1990 model year vehicles.  In addition,
all fuel modifications considered below to reduce NOx emissions
were evaluated with fuels already containing 2.1 weight percent
oxygen.  Hence,  it is clear that significant reductions in NOx
can be achieved without lessening the oxygen requirement.  In any
case, section 211(k)(2) allows EPA to waive the oxygen content
requirement for reformulated gasoline only if the Administrator
determines that compliance with the oxygen content requirement
would prevent or interfere with attainment of an applicable
NAAQS.   EPA therefore could not waive this requirement for the
reasons presented by the commenter.

     Many comments addressed the cost-effectiveness of NOx
control.  Some raised questions about the accuracy of the costs
used, which impacts the cost-effectiveness determination.  The
cost-effectiveness analysis presented later in this section uses
improved costs and is believed to be representative of the impact
of NOx control on the average refinery.  Some of the comments
supported NOx control on the basis of the cost-effectiveness
analysis presented with the proposal,  citing the benefits that
specific locations would realize by reducing the mobile source
contribution to the NOx inventory.  As discussed below, although
the benefits of NOx reduction vary from city to city,  in general
most ozone nonattainment areas will benefit from a reduction in
NOx in addition to the mandated reductions in VOC emissions.
                               303

-------
     Some comments questioned the validity of the information
which claims ozone benefits from NOx reductions,  while others
supported the benefits of reducing NOx emissions  in ozone
nonattainment areas.  Most comments centered around the question
of how much a given nonattainment area would benefit from NOx
reductions in addition to VOC reductions.   The following section
presents EPA's response to these comments.

2.    Benefits of NOx Reductions

     EPA proposed NOx standards based on the reduction in ozone
formation that is expected to result from a combination of
reductions in both VOC and NOx emissions.   Gasoline vehicles
contributed 20-35% of total urban NOx emissions in 1990,  and are
expected to continue to represent a large fraction of the NOx
inventory in the year 2000.54  Hence, control of mobile source
NOx emissions can be an important step in meeting the air quality
requirements of various urban areas.

     A recent report by the National Research Council (NRC)
concluded that "the optimal set of controls relying on VOCs,  NOx,
or, most likely,  reductions of both, will vary from one place to
the next."55  In response to the NRC report, EPA acknowledged
that NOx control in addition to or instead of VOC control is
likely to reduce ozone in many areas.56  Similarly, EPA
acknowledges that the magnitude of the ozone benefits for VOC
controls is dependent on the particular area.57

     Many stationary source emission control measures can be
applied on a selective basis to individual areas  based on their
air quality needs.  For ozone control strategies  based on fuel
controls, however, it is extremely difficult if not impossible to
     54Although by  the year  2000  a portion of the vehicles will
have lower NOx emissions because  they will  meet  the  "Tier 1"
emission standards, these emissions benefits are expected to  be
offset by increases in annual vehicle miles travelled.

     "National Research  Council,  Rethinking the  Ozone Problem
in Urban and Regional Air Pollution, National Academy Press,
Washington, B.C.,  1991.

     56U.S. EPA,  "Summary of Public Comments and  EPA Responses
on the Draft Report 'The Role of  Ozone  Precursors in
Tropospheric Ozone Formation and  Control'," EPA-454/R-93-025,
July 1993.

     57For example, due to the transport of  ozone over the
region, control strategies  that might not benefit one area
within the nonattainment region may help  an area downwind.

                               304

-------
tailor the relative reductions in VOC and NOx emissions to the
specific needs of individual areas.  The fungible nature of the
gasoline distribution system in this country would simply not
allow for detailed specific VOC and NOx standards for each and
every individual area.  Such a requirement would greatly hamper
the efficient distribution of the fuel, thereby resulting in
extremely large increases in fuel distribution costs, rendering
such emission controls through fuel reformulation cost
ineffective.  The reformulated gasoline standards avoid the vast
majority of these concerns by applying one consistent set of fuel
standards to a range of areas.  By so doing, the program achieves
significant ozone reductions on average in a cost effective
manner,  while at the same time recognizing that the controls
would be more effective at reducing ozone in some areas than in
others.

     With this in mind, EPA proposed in February 1993, that a NOx
reduction standard be applied to all RFC areas.  While either NOx
or VOC reductions by themselves may provide little ozone benefit
in a minority of the covered areas, the combination of NOx
control  with VOC control will ensure that significant ozone
reductions are achieved in all of the covered areas.  Applying
fuel NOx controls nationwide  (as is done out of necessity with
tighter vehicle emission control standards)  would be
inappropriate, however, since most of the additional NOx
reductions would be achieved in areas for which little or no
ozone benefit would result.  Therefore, EPA believes it
appropriate to apply a NOx reduction standard to the same areas
for which the reformulated gasoline standards apply.

     The following presents a summary of the scientific evidence
which favors NOx reductions for most, if not all, ozone
nonattainment areas.  The benefits include the primary benefit of
reducing ozone precursors as well as secondary benefits which
will benefit human health and the environment.

a.   Effect of NOx on Ozone

     Although both NOx and VOC emissions are regulated for
certain types of sources, including motor vehicles, the primary
focus of past ozone abatement strategies has been on VOC.  This
is largely due to the belief that the most cost effective way to
reduce ozone formation was to limit the amount of VOC in the
ambient  air.  However, many areas where VOC reduction has been
applied as the primary ozone control strategy have not reached
attainment.  In recent years, improved knowledge of emission
sources  and atmospheric chemistry has shown that both VOC and NOx
emissions can enhance ozone formation.  The benefits of either
VOC or NOx emission reductions can depend on the ratio of VOC to
                               305

-------
NOx in the airshed.58  This ratio varies spatially and temporally
within a given air basin and can differ between severe ozone
episodes or more typical ozone exceedances.  However,  although
simple smog chamber experiments typically show that combined VOC
and NOx reductions are less effective at controlling ozone than
reducing either one alone,  the complexity of variations in
meteorology,  emissions sources, and control feasibility often
make combined VOC and NOx control the option of choice in real
world situations.59

     Prior to the development of more sophisticated models,
information about the response of ozone levels to VOC and NOx
emissions was often based on EKMA modelling.  This modelling
produces graphical representations (isopleths) of the
relationship between VOC and NOx concentrations in the formation
of ozone.  Regions where ozone formation is either VOC or NOx
limited are easily determined from these plots.  At low VOC to
NOx ratios, less than about 8 on a molar basis, ozone formation
is VOC limited, so VOC control is more effective.  A low ratio is
typical of highly urbanized areas.  For VOC to NOx ratios greater
than 8, ozone formation is NOx limited, and NOx control is more
effective.  For instance,  the Houston area has been identified by
the EKMA analyses as an area in which ozone formation is NOx
limited and thus VOC control alone would do little to reduce
ozone levels in this area.60  Many areas downwind of urban
centers also have large VOC to NOx ratios and so would be NOx-
limited.

     Ozone grid-based modelling studies using more sophisticated
analytical techniques have shown that NOx control is generally a
more effective ozone reduction strategy than VOC control.   The
recent National Research Council study61 utilized work in the
California South Coast basin62 to suggest that NOx control is
     58Finlayson-Pitts,  B.J.  and J.N.  Pitts,  Jr.,  "Atmospheric
Chemistry of Tropospheric Ozone  Formation:   Scientific and
Regulatory Implications,"  Air  & Waste,  Vol.43, August,
pp.1091-1100, 1993.

     59U.S. EPA,  EPA-454/R-93-025 .

     "National Research Council, Rethinking the Ozone Problem
in Urban and Regional Air Pollution,  National Academy Press,
Washington, B.C., U.S.A., 1991.

     "Ibid.

     62Milford, J.B.,  Russell, A.G.,  and McRae, G.J., "A New
Approach to Photochemical Pollution  Control:  Implications of
Spatial Patterns in Pollutant Responses  to  Reductions in

                               306

-------
essential to meeting the ambient air quality standards and that
even marginal NOx reductions result in improved air quality in
most areas.  Subsequent studies63'64'65 have shown that NOx control
is an effective ozone reduction strategy for the northeast United
States,  including the Baltimore, Philadelphia,  New York-
Connecticut, and Boston-Maine regions.  The EPA ROMNET report on
ozone modelling for the northeastern states indicated that NOx
control reduced ozone for nearly all of the northeastern
modelling domain, except New York City and nearby areas.  In New
York City,  ozone levels increased when only NOx emissions were
reduced, indicative of VOC-limited ozone formation.  When both
VOC and NOx emissions were reduced, however, ozone was reduced
throughout the modelling domain.  Of course, with Phase II
reformulated gasoline, NOx emission control will only occur along
with significant VOC emission reductions (several times greater
than the NOx reductions).   Therefore, ozone reductions are
expected over the entire region modelled by the EPA study, which
includes four of the mandated nine reformulated gasoline regions.
These include areas that are VOC-limited as well as NOx-limited
for ozone formation as well as most of the opt-in areas.  The
ROMNET study also suggested that the benefits of NOx controls may
extend beyond rural regions since net ozone transport into cities
is reduced.

     Similar modelling has been performed for the Lake Michigan
region  (including Chicago, IL; Muskegon, MI; and Milwaukee,
WI) .66  The results of this study indicated that NOx control
without VOC control could be detrimental for the Lake Michigan
Nitrogen
Oxides and Reactive Organic Gas  Emissions,"  Environ.  Sci.
Techno!., v 23, pp. 1290-1301, 1989.

     63Sillman,  S.,  Samson,  P.J.,  and Masters,  J.M., "Ozone
Production in Urban Plumes Transported over  Water:
Photochemical Model and Case  Studies  in the  Northeastern and
Midwestern United States," J. Geo.  Res.,  v.  98,  n.  D7,  pp.
12,687-12,699,  1993.

     64Sillman,  S. and  Samson,  P.,  "Simulated Ozone over Lake
Michigan and the Northeast Corridor:  Identifying the
Differences between Hydrocarbon-limited and  NOx-limited
Regions," Presentation at the 86th  Annual Meeting and
Exhibition, Denver, CO, June  13-18,  1993,  93-WP-68B.05.

     65U.S. EPA, Regional  Ozone Modeling for  Northeast Transport
(ROMNET), EPA Report 450/4-91-002a,  June 1991.

     66Sillman and Samson, 86th Annual Meeting of Air & Waste
Management Association, June  13-18,  1993.

                               307

-------
region.  As long as VOC reductions are greater than NOx
reductions, however, the combination of the two will result in
ozone benefits.  The sensitivity of the modelling results to the
VOC emissions inventory was also tested, and showed that if the
VOC emissions inventory was underestimated (as discussed in the
next paragraph),  NOx emission reductions become universally
effective in the Lake Michigan region.

     The NRC and EPA67 have indicated that emissions inventories
of VOC and carbon monoxide may have been underestimated in the
past.  The estimated relative benefits of VOC or NOx control as
ozone reduction strategies depends upon the assumption of the
emission inventories.  As the VOC emission inventory estimate
increases, NOx control becomes a more effective strategy for
reducing ozone levels.  Several changes have led to higher
estimates in the VOC emission inventory since the grid models
cited in the previous paragraphs were performed.  EPA's MOBILES
emission model estimates significantly higher VOC emissions than
the MOBILE4 model which was used in the ozone modelling studies
cited above.  In addition, indications are that the effectiveness
of stationary source controls is not as great as assumed in the
ozone modelling studies.68  This also increases the VOC emissions
inventory estimate.  Therefore, even though NOx control has been
shown to be very effective in reducing ozone, the efficacy of NOx
reduction may have been underestimated in past modelling.

     Ambient measurements have also indicated that the VOC
inventory may be underestimated.  Measurements of VOC/NOx ratios
have been typically a factor of approximately two higher than
predicted from the emission inventories in major cities
throughout the U.S.69'70  The measurement of ambient VOC may be
actually too low because measurement methods tend to
     67U.S. EPA,  EPA-454/R-93-025,  July 1993.

     ^Statement  by  Bill  Oliver,  Radian Corp.,  Presentation at
the 1992 Conference on Lake Michigan  Ozone and the Clean Air
Act, Chicago, IL, September 1992.

     69Finlayson-Pitts  and Pitts,  Air  & Waste,  Vol.43, August,
pp.1091-1100, 1993.

     70Fujita, E.M., Croes,  B.E.,  Bennett,  C.L., Lawson, D.R.,
Lurmann, F.W., and Main,  H.H.,  "Comparison of  Emissions
Inventory and Ambient Concentration Ratios of  CO,  NMOG, and NOX
in California's  South Coast Basin," JAPCA.  v  42,  pp.264-276,
1992.

                               308

-------
underestimate the total.71  The VOC/NOx ratio of most urban areas
was thought to be much lower than ambient measurements would
indicate.  Several other studies72'73'74 have concluded through a
technique called source/receptor modelling that the VOC emission
inventory is underestimated.  If the VOC emission inventory is
severely underestimated, then regions where NOx control are
estimated to increase ozone levels may not exist.

     Another concern is that biogenic emissions or their
importance may have been underestimated in the past.75  In the
Atlanta area, biogenic emissions of VOC may be high enough that,
in the absence of NOx control,  even 100% control of anthropogenic
VOC would not enable the region to meet the National Ambient Air
Quality Standard (NAAQS) ozone standard.76  In addition, the VOC
emissions from biogenic sources tend to be more reactive than
those from stationary or mobile sources.  Results of EPA's ROMNET
study suggest that the impact of biogenics on ozone formation
extend beyond the southeastern U.S. to include much of the
northeast, as well.  To the extent that biogenic VOC emissions
compound the air quality problems of a given region, NOx control
becomes more effective.
     71 Chameides, W.L.,  Fehsenfeld,  M.O.,  Rodgers,  M.O.,
Cardelino, C., Martinez, J.,  Parrish,  D.,  Lonneman,  W.,  Lawson,
D.R., Rasmussen, Zimmerman, P., Greenberg,  J.,  Middleton,  P.,
and Wang, T., "Ozone Precursor Relationships  in the  Ambient
Atmosphere," J. Geo. Res., v. 97, n.  D5, pp.  6037-6055,  1992.

     72 O'Shea, W.J.  and  Scheff,  P.A.,  "A Chemical Mass Balance
for Volatile Organics in Chicago," JAPCA,  v.  38,  pp.  1020-1026,
1988.

     73 Aronian, P.P., Scheff, P.A.,  Wadden, R.A., "Wintertime
Source-Reconciliation of Ambient Organics," Atmospheric
Environmental, v. 23, n. 5, pp.  911-920, 1989.

     74 Harley R.A.,  Hannigan, M.P.,  and Cass  G.R.,
"Respeciation of Organic Gas  Emissions and the  Detection of
Excess Unburned Gasoline in the Atmosphere,"  Environ.  Sci.
Technol.. v. 26, pp. 2395-2408,  1992.

     75Chameides, W.L., Fehsenfeld, M.O.,  Rodgers, M.O.,
Cardelino, C., Martinez, J.,  Parrish,  D.,  Lonneman,  W.,  Lawson,
D.R., Rasmussen, Zimmerman, P., Greenberg,  J.,  Middleton,  P.,
and Wang, T., "Ozone Precursor Relationships  in the  Ambient
Atmosphere," J. Geo. Res.. v. 97, n.  D5, pp.  6037-6055,  1992.

     76At  this time  Atlanta  is not participating in the
reformulated gasoline program.  However, the  Atlanta area  is a
serious ozone nonattainment area.

                               309

-------
     Areas downwind of urban plumes tend to have high VOC/NOx
ratios indicative of NOx-limited ozone formation.  This may be
due to increases in local biogenic emissions or transformation
and subsequent deposition of nitrated compounds created upwind.
Reducing the upwind NOx emissions may be the only means for
meeting the ozone standard downwind,  even if those emission
reductions do not improve air quality in the upwind areas.  Many
nonattainment areas are downwind of other nonattainment areas,
particularly in the Northeast corridor and the Lake Michigan
region.  Hence, NOx emissions can impact more than just emission
source areas.

     Since reformulated gasoline can assist control of both VOC
and NOx emissions from motor vehicles, its use should provide
ozone reductions in areas that are either VOC or NOx limited.  In
the VOC limited areas where NOx emission reductions may cause
ozone increases, the magnitude of VOC reductions provided by
reformulated gasoline should prevent any ozone increases  (as
demonstrated by the Lake Michigan and other studies).   In the
great majority of areas, reformulated gasoline with NOx control
will provide reductions in ozone.  As such, the RFG program
represents an ideal opportunity to reduce ozone in all areas
through both VOC and NOx control.

b.   Secondary Benefits

     The secondary benefits of NOx emission reductions are also
important.  As discussed below, these include reductions in acid
rain, deposition of nitrates in lakes and estuaries,  ambient
levels of nitrogen dioxide, secondarily formed particulates, and
highly toxic nitrated polycyclic organic matter, as well as
potential global warming benefits.

     NOx emissions from gasoline vehicles are a significant
fraction of the overall NOx emission inventory.  While the
primary cause of acid rain may be sulfur dioxide emissions, the
benefit of reducing NOx emissions should not be ignored.  NOx
emissions can be precursors to acid aerosols in the form of
nitrates or by contributing to the formation of sulfates by
reaction with sulfur dioxide gas.  Hence, any reduction in NOx
emissions from gasoline vehicles should help reduce acid rain
precursors.  (Furthermore, the acid rain benefits of NOx
reductions through gasoline reformulation will be enhanced by the
fact that the majority of the NOx emission reductions are
expected to be achieved through the removal of sulfur from the
gasoline.)

     Several studies indicate that atmospheric deposition
accounts for a significant portion of nitrogen deposited into
certain sensitive estuaries (e.g., 25% of the nitrogen in
                               310

-------
Chesapeake Bay is due to atmospheric deposition).7V  The
atmospheric deposition of nitrates can also contribute to the
depletion of oxygen in lakes and estuaries as a result of an
increase in the algae and other plant matter growth in these
waterways.  The benefit of reducing these nitrates through
control of gasoline vehicle emissions has not previously been
evaluated, but could be significant.

     Nitrated particulate matter related to NOx emissions (e.g.,
ammonium nitrate) has also been shown to be a significant
fraction of some area's PM-10 problem.78  Some particulate
nitrates may be toxic, mutagenic,  or carcinogenic.79  These small
particles are carried deep into the lung and are known to cause
potentially serious respiratory effects.  They also contribute to
impaired visibility, raising the public's awareness of this air
pollution.  Any NOx reduction should reduce these particles.
Since mobile source NOx emissions are ground-based and
distributed over wide areas, reducing these could have more
impact on human health than reductions from stationary sources.

     EPA has also cited polycyclic organic matter (POM)  as an
air toxic.80  The toxicity of these compounds increases greatly
when nitrated.  A reduction in NOx emissions should reduce the
transformation of POM to the more toxic nitrated form.

     A change in NOx emissions is also known to affect
concentrations of the hydroxyl radical in the troposphere (the
lower layer of the atmosphere).   The concentration of the
hydroxyl radical in turn impacts the concentration of
tropospheric ozone, a greenhouse gas.  NOx reductions thus could
have positive benefits in reducing global warming.  However,  the
impacts of a reduction in U.S.  mobile source emissions of NOx on
     77"Progress Report  of  the Nutrient Reduction Reevaluation,"
Chesapeake Bay Program, February  1992.

     78Farber, R.J.,  Baas,  J.R.,  Pilinis,  C.,  and Countess,
R.J., "Modelling the Secondary Component  of  PM10, " PM1; Standards
and Nontraditional Particulate Source  Controls,  Chow,  J.C. and
Ono, D.M. editors, Air  and Waste  Management  Association,
Pittsburgh, PA, November,  1992.

     79California Air Resources Board,  "The Effects of Oxides of
Nitrogen on California  Air Quality," Report  No.  TSD-85-01,
March 1986.

     80U.S. EPA, Motor Vehicle-Related  Air Toxics Study.  EPA
Report 420-R-93-005, April 1993.

                               311

-------
global tropospheric ozone concentrations and the corresponding
relationship to global warming are unclear at this time.81

     In sum, while the benefits of reducing ozone levels as a
result of NOx reductions are the primary basis for the controls
being promulgated, the other direct and indirect benefits of
reducing atmospheric NOx levels also appear to be substantial.

3.   Consideration of §202 Motor Vehicle Controls

     Before controlling or prohibiting a fuel or fuel additive
under §211(c)(1)(A), the Administrator must consider "other
technologically or economically feasible means of achieving
emission standards under section  [202]."  This has been
interpreted as requiring consideration of regulation through
motor vehicle standards under §202 prior to regulation of fuels
or fuel additives under §211(c)(1)(A)  [Ethyl Corp. v.
Environmental Prot. Agcy., 541 F.2d 1,  32 (B.C. Cir. 1976)].
This does not establish a mandatory preference for vehicle
controls over fuel controls, but instead calls for the good faith
consideration of motor vehicle standards before imposition of
fuel controls [541 F.2d at 32 n.66].   This reflects Congress'
recognition that fuel controls under §211(c)(1)(A) might
logically involve controls on fuel composition itself,  while
vehicle standards under §202 are generally performance standards,
regulating vehicle emissions and not the design or structure of
the vehicle.  Fuel controls might therefore lead to greater
government involvement in the regulation of the manufacturing
process than would be expected from vehicle controls [541 F.2d at
11 n.13] .

     Congress addressed this concern by requiring agency
"consideration" of vehicle standards under §202 before imposition
of fuel controls under §211(c)(1)(A).   It is important to note
that the Administrator must in good faith consider such vehicle
controls,  but retains full discretion in deciding whether to
adopt either fuel or vehicle controls,  or both [541 F.2d at 32
n.66] .

     In evaluating motor vehicle controls under §202 in this
context,  the first major point to consider is that EPA has
already imposed more stringent NOx control standards on motor
vehicles.   The Tier 1 standards for light-duty motor vehicles and
trucks require reductions in light-duty motor vehicle NOx
emissions starting with model year 1994, with a percentage phase-
in of the more stringent Tier 1 standards until they apply to all
     81Hangebrauck,  Robert  P.  and John W.  Spence, U.S. EPA,
"Ozone and Global Warming," August,  1992.

                               312

-------
new model year 1996 and later light-duty vehicles and trucks.82
In-use compliance with these standards is also phased-in, such
that model year 1998 and later new light-duty vehicles must
comply with the Tier 1 NOx standard in-use.  The same applies for
light-duty trucks, with model year 1999 instead of 1998  [56 FR
25787]  For heavy-duty vehicles, EPA recently reduced the NOx
standard to 4 g/bhp-hr, starting with model year 1998  [58 FR
15781, March 24,  1993]

     While these motor vehicle and motor vehicle engine controls
are expected to reduce mobile source emissions of NOx, this
result is limited by certain basic facts.  First, the standards
only apply to new motor vehicles and engines. It will therefore
take several years after the first model year of the standards
before vehicles and engines certified  to these standards will
make up a significant portion of the motor vehicle fleet.83  In
addition, it is expected that emissions reductions based on the
reduction in the NOx standard will be offset to a significant
extent by an increase,  over time, in total vehicle miles
travelled.84

     In addition to motor vehicle controls under §202, EPA has
recently adopted or proposed other controls aimed at in-use NOx
emissions from mobile sources.  The enhanced inspection and
maintenance (I/M)  rules call for use of these more stringent I/M
procedures starting with 1996 [57 FR 52950, November 5, 1992].
EPA has also proposed standards that would limit NOx emissions
from new large horsepower diesel non-road engines, pursuant to
§213 of the Act [58 FR 28809, May 17, 1993].   While enhanced I/M
programs will directly affect the whole motor vehicle fleet, the
non-road engine regulations are similar to the motor vehicle
regulations under §202 in that they would apply to new non-road
engines only,  and therefore involve a certain time before a
significant portion of this category of non-road engines is
replaced by new engines certified to meet the NOx standards.
Finally, the centrally fueled fleets program and the California
Pilot program  [§§246 and 249] will result in tighter NOx
standards for a small segment of the motor vehicle fleet.
     8256 FR  25724,  June  5,  1991.   Also,  note that the Tier 1
standards apply to  light-duty  trucks  with a loaded vehicle
weight rating of 3,750 Ibs.  or less.

     83Fleet  turnover  is  included in EPA's MOBILESa model  [58 FR
29409, May 29, 1993] .  As of 1990,  the median vehicle age was 7
years  (i.e., one-half of the vehicles were younger than this
and one-half were older).

     84Ibid.

                               313

-------
     Additional mobile source controls, whether under §202 or
under other authority such as described above, may well be cost
effective and reasonable options that EPA might decide to adopt.
However, there are certain limitations imposed by Congress on the
adoption of more stringent standards.  For example, Congress
spelled out when and under what conditions EPA may promulgate
more stringent NOx standards for light-duty vehicles and trucks.
Congress required that EPA conduct a study on whether more
stringent standards for light-duty vehicles and trucks should be
adopted, and report back to Congress no later than June 1, 1997
[§202(ij(1),(2)].   Based on the study EPA must conduct a
rulemaking to determine whether there is a need for such further
reductions, whether the technology will be available for such
reductions, and whether further reductions in emissions from such
vehicles will be cost effective.  If these determinations are
made in the affirmative, then EPA would proceed to promulgate
emissions standards that are more stringent than the Tier 1
standards  [§202(i)(3)(C)].   If EPA does promulgate more stringent
standards, they may not take effect any earlier than model year
2004, and no later than model year 2006.

     It is clear from this that EPA has not,  at this time,
completed the lengthy process for determining whether or not more
stringent standards should be established for light-duty vehicles
and trucks under §202(i).   Congress established a detailed
provision spelling out the procedures to follow and the
substantive determinations that must be made before such controls
could be adopted.   There is no indication, and EPA does not
believe, that these mandated procedures and criteria preclude the
exercise of discretion under §211(c)(1)(A) prior to completion of
the rulemaking under §202(i) .  Congress required that EPA
consider motor vehicle controls, but did not establish a
mandatory preference for such controls and did not preclude the
adoption of fuel control prior to a decision on Tier 2 motor
vehicle standards.

     In any case,  it is clear that a decision to impose more
stringent NOx standards for light-duty vehicles and trucks under
§202(i) could not take effect prior to model year 2004.  It would
then take several years before a significant portion of the in-
use fleet would include vehicles or trucks certified to a NOx
standard more stringent than the Tier 1 standard.  A similar
situation would apply to a more stringent NOx standard for heavy-
duty engines.   The mandatory leadtime and stability provision of
§202(a)(3)(C)  would preclude imposition of more stringent NOx
standards for heavy-duty engines until model year 2001 at the
earliest.   It would again take several years before a significant
portion of the in-use heavy-duty fleet contained engines
certified to a more stringent NOx standard.  For non-road engines
and vehicles,  EPA expects  to continue to explore NOx controls for
this category of mobile sources.  But as with motor vehicles, any
new or more stringent NOx standards will only apply to new non-

                               314

-------
road engines, after providing a reasonable period for leadtime.
The effect on in-use emissions is delayed based on the time
needed before new non-road engines replace earlier models.

     Given these circumstances, there are several important
reasons why promulgation of a NOx reduction standard for
reformulated gasoline is important,  whether or not additional
vehicle or engine controls are later adopted by the Agency.
First, emissions reductions from the NOx performance standard
would start as soon as the standard is applicable, with no delay
based on fleet turnover time.  Significant NOx emission
reductions would be achieved right away, in the spring of 2000,
while more stringent light-duty or heavy-duty standards would not
be expected to significantly affect in-use emissions until much
later in that decade.  Second, a NOx reduction standard for
reformulated gasoline would act to reduce emissions from all
mobile sources that use gasoline, whether highway or non-road,
while §202 or §213 standards only act to limit emissions from new
engines or vehicles in that specific category of mobile sources.
Third, this fuel control is specifically aimed at areas of the
country that are in nonattainment for ozone, and is limited in
time to that part of the year when ozone is of most concern.
Vehicle or engine controls, in contrast, apply to all new engines
or vehicles, wherever they are used, throughout the year.  This
fuel control thus allows a more narrow regulatory solution aimed
at the specific geographical areas and time periods when control
is needed.  Fourth, the expected increase in vehicle miles
travelled over time leads EPA to believe that this fuel control
is needed to continue to achieve the in-use NOx emission
reductions necessary for many areas of the country to reach
attainment for ozone.  Fifth, the NOx controls required in this
rule will reduce NOx emissions at a reasonable cost, about $5,000
per ton.  This compares favorably with the cost-effectiveness of
other mobile and stationary source control strategies, as
supported in section H below.  Finally,  the NOx fuel standard
adopted here minimizes any concern there might be that a fuel
control would tend to interfere in the production process by
directing refiners on how to make their product.  The NOx
standard is not a fuel recipe, but instead establishes a
performance standard, leaving refiners free to produce their
gasoline in any way that achieves the desired reductions.

     EPA is not at this time determining whether additional
vehicle or engine NOx controls should be adopted under §202 or
any other provision of the Act.  Instead, based on all of the
above, EPA believes that a NOx reduction standard for
reformulated gasoline under §211(c)(1)(A) is an appropriate
exercise of discretion, whether or not the agency imposes
additional vehicle or engine NOx controls in the future.
                               315

-------
C.   Derivation of Fuel Parameter Control Costs

     This section describes the methodology used to determine the
costs of controlling individual fuel parameters and highlights
important factors which influence the estimation of fuel
parameter control costs.  Also addressed are the fuel economy
impacts of different fuel formulations and how they were included
in the analysis.  A table summarizing the costs derived from this
analysis can be found in subsection 5 below.

1.   Refinery Modelling and Assumptions

     Refinery modelling is the best method currently available
for estimating the cost of particular refinery changes, such as
those which are required to produce fuel formulations which meet
various emission performance standards.  Mathematical refinery
models are created to estimate the changes in manufacturing costs
associated with the control of one or more fuel components.
Optimal solutions are found by the model such that all product
specifications, for gasoline as well as other refinery products,
are met at the lowest cost.  The change in manufacturing costs,
or the incremental cost of control, is the difference between the
cost of producing fuel under a defined base case and the new cost
which occurs when a fuel specification is changed.

     EPA received many comments on the costs used to develop the
cost-effectiveness analysis which supported the proposed Phase II
RFC standards.  EPA used refinery modeling studies as the basis
for the proposed Phase II costs.  One comment suggested that the
costs used in the analysis did not accurately reflect all of the
costs incurred, since they were only based on manufacturing
costs.  In particular, if the fungibility (the ability of a
mixture of two complying fuels to still comply with the
standards) of Phase II fuel became an issue, then the costs would
be significantly higher than those used here due to the increased
complexity of the distribution system.  Since EPA has
demonstrated  (see section III) that fungibility will not be a
problem for either Phase I or Phase II reformulated gasoline,
such costs do not need to be included in this analysis.  An
exception to this are fuels containing ethanol, since such fuels
must be segregated from non-ethanol containing fuels.  However,
this segregation has been clearly required throughout all EPA's
proposed rules regarding reformulated gasoline.  Also, the
refinery studies used below either use MTBE-based gasolines not
requiring segregation, or use MTBE and ethanol-based fuels with
ethanol prices sufficiently high to cover the costs of
segregation.  Thus, no additional modifications to the analysis
are necessary.

     Another comment suggested that it is inappropriate to use
independent refinery models to generate costs for controlling
individual parameters, and that aggregate nationwide models

                               316

-------
should not be used.  Rather, costs should be generated from
refinery models for regions with high levels of RFC production,
such as PADD I.  EPA disagrees.  The actual cost of producing
reformulated gasoline at various NOx emission performance levels
will include the costs for regions having high reformulated
gasoline concentrations and those with low concentrations.  Both
need to be considered if the costs developed here are to be
comparable to those of other NOx control programs.  The costs
imposed on individual regions is important when competitive and
regional economic impacts are evaluated.  However, national
average costs should be used when comparing the cost-
effectiveness of the reformulated gasoline program to that of
another program.

     A third comment suggested that the costs estimated in this
analysis should include the costs of complying with the program,
not just the manufacturing costs.  This is true.  However,
changing the emission performance standards for Phase II should
have little or any effect on the non-manufacturing costs of the
program, except for fuel economy effects, which are addressed
below.  The costs of recordkeeping and other compliance related
costs should be the same for Phase II as they were for Phase I
and not depend on the level of the Phase II standards.  Thus,
they do not change in this incremental analysis and need not be
included.

     Two basic methodologies can be used to estimate the cost of
various RFC standards.  In one, a version of EPA's complex model
for reformulated gasoline is expressed as a series of linear
equations and integrated into a refinery model.   The refinery
model is then used to determine the lowest cost of reducing
emissions of the particular pollutants of concern (VOC, NOx and
toxics).  This method is preferable when the emission effects of
fuel changes are known and when sufficient time is available to
integrate this emission information with the cost analysis.  The
weakness of this approach is that the results become suspect
whenever the emissions model is changed.  Since the refinery
model was trading off up to eight fuel parameters at a time based
on the coefficients of the emissions model, changes to these
coefficients can lead to markedly different results.

     The other method more loosely couples the emissions model
and the refinery model.  The refinery model is used to develop
costs for varying individual fuel parameters.  The effect of
these fuel changes on emissions and their cost are then compared
and the most cost-effective fuel modifications identified.  The
refinery model is then rerun, if necessary, to first modify the
most cost-effective fuel parameter, then the second, and so on.
This method always results in a less optimum refinery
configuration  (i.e., higher costs) than the first method, because
the refinery model is only evaluating a small number of possible
ways to achieve the specified emission reduction.  Of course,

                               317

-------
care must be taken to ensure that emissions of the other
pollutants of interest are being maintained or reduced while
control of the primary pollutant is being evaluated.

     EPA, for its own work (and the work contracted by the
Agency),  chose to follow the second method for both time and
resource related reasons.   The complex model was being developed
concurrently with the cost analysis.  Repeatedly modifying the
refinery model to incorporate each new version of the complex
model would have been costly and time consuming,  and would have
delayed the promulgation of the standards.  As described below,
as fuel modification costs using the first method have become
available from other sources (only very recently),  the parameter
by parameter costs were compared to complex model based costs to
ensure basic consistency in the two estimates of how refineries
would act to reduce fuel-based emissions.

     The majority of the refinery modelling on which EPA
estimated its cost estimates was performed by Bonner and Moore (B
& M)  through an EPA contract with the Southwest Research
Institute and a Department of Energy (DOE) contract with the
National Institute for Petroleum and Energy Research (NIPER) .85
Bonner and Moore, for EPA, created a specialized version of their
refinery model to estimate the cost of the RFC program.   In
addition to the work done by B & M, EPA staff ran the B & M model
to determine costs for two of the parameters,  as discussed
further below.86  Other refinery modelling and fuel modification
costs used or considered were generated by the Department of
     85Bonner and Moore Management  Science,  "Study of the
Effects of Fuel Parameter Changes  on  the  Cost  of  Producing
Reformulated Gasoline," Prepared for  EPA  under contract  through
Southwest Research Institute and the  National  Institute  for
Petroleum and Energy Research.  This  data was  made  available to
the public through the following document:   "DOE  and API Phase
II Cost Estimates," EPA Memorandum from Lester Wyborny,  FSSB,
to the Air Docket, November 4, 1993.  The Bonner  and Moore work
performed for this analysis was based on  a  revised  version of
the model used to generate costs for  the  draft RIA.   This new
version includes provisions to prevent dumping of fuel
components into conventional gasoline.

     86Lester Wyborny, U.S. EPA,  "Aromatics  and E200
Reformulation Costs," Memorandum to the Air Docket,  December
10, 1993.

                               318

-------
Energy through the Oak Ridge National Laboratory,   and for API,
WSPA,89, Auto-Oil90 and the National Petroleum Council91 by  the
Turner, Mason and Company (Turner/Mason).   A more complete
comparison of EPA's analysis with the other refinery analyses may
be found in section G below.

     EPA used B & M-estimated Phase II RFG sulfur and olefin
costs in the February 26, 1993 proposed rule.   That B & M data,
like the data used here for the final rule, is most attractive to
EPA because unlike other modelling work on the Federal RFG
program, it attempts to model the effect of the RFG program on
the entire refining industry.  However, commenters pointed out
that the B & M modelling work for the proposed rule permitted
parameter values to increase in conventional gasoline (a
violation of the antidumping restrictions), which would lead to
low estimated investments and decreased RFG program costs.  EPA
corrected that situation in the final rule by directing B & M to
cap parameter levels in conventional  gasoline at 1990
conventional gasoline levels.  This approach,  while it prevents
any shifting of regulated gasoline components to conventional
fuel, is overly restrictive in that most refiners will use the
complex model to evaluate their conventional gasoline composition
after 1994 to compare it to their 1990 gasoline.  The complex
model would give refiners more flexibility in meeting their
antidumping restrictions resulting in some cost savings.   Thus,
the B & M modelling work for the final rule would tend to be
slightly overstated.

2.    Factors Influencing Cost

     A number of economic reporting factors influence the
estimated cost of producing gasoline  in addition to the operating
     87Lester Wyborny,  U.S.  EPA/FSSB,  "DOE and API Phase II Cost
Estimates," Memorandum to the Air Docket,  November 4,  1993.

     88Ibid.

     89Turner, Mason  &  Company,  "WSPA Study of the Cost Impacts
of Potential CARB Phase 2 Gasoline  Regulations,"  Prepared for
the Western States Petroleum Association,  November 18,  1993.

     90Turner, Mason  &  Co.,  "Costs  of Alternate Gasoline
Reformulations, Results of  U.S. Refining  Study,"  Prepared for
the Economics Committee of  the Auto-Oil Air Quality Improvement
Research Program, April 1992.

     91National  Petroleum  Council,  "U.S.  Petroleum Refining:
Meeting Requirements for Cleaner Fuels and Refineries," August
1993.

                               319

-------
costs of a specific refinery.  Those factors which impacted EPA's
cost analysis are discussed below.

a.   Rate of Return on Investment

     The internal rate of return is an economic factor used by
businesses such as the refining industry in evaluating the
economic breakeven point for their investments.  Although the
refining industry and their consultants often evaluate potential
investments at a 15 percent real rate of return,  EPA has
determined that the historical rate of return for the oil
industry is much closer to 10 percent.92  This appears to reflect
the fact that most pre-investment analyses are inherently
optimistic and do not quite yield their predicted profits.  To
reflect both the pre-investment actions of refiners and real
results, EPA believes it most appropriate to operate the refinery
model with a 15 percent real rate of return hurdle on capital
investment to most accurately reflect the tradeoff between
operating costs and capital.  Once the refinery model has
determined an optimum processing approach,  capital costs are then
reannualized using a 10 percent real rate of return to reflect
the actual and historically acceptable performance of capital.
This is very similar to the approach taken by the Auto-Oil group
in their refinery modeling,93 where they used a 15 percent real
rate of return internal to the refinery model and then present
the costs using both 9 and 13 percent real rates of return.

     In adjusting annualized capital costs for the parameter
changes described below, a 0.171 annual cost recovery factor
(i.e.,  the fraction of total capital investment that must be
recovered each year to achieve the specified real rate of return
on capital) was used in lieu of the 0.226 cost recovery factor
contained in the Bonner and Moore model.94  In addition, where
refinery modeling data were extracted from other studies for
direct use or for comparison purposes, the cost recovery factors
were adjusted using the same methodology.

b.   Cost Amortization
     92U.S. EPA,  "Analysis  of  the Economic and Environmental
Effects of Methanol as an  Automotive  Fuel,"   Special Report,
Office of Mobile Sources,  September 1989.

     93Turner, Mason &  Co. ,  April 1992.

     94U.S. EPA,  "Control  of Sulfur and Aromatics Contents of
On-Highway Diesel Fuel," Draft  Regulatory Impact Analysis,  July
1989.

                               320

-------
     Phase II VOC and NOx emission controls are necessary only
during the high ozone period.  Hence,  any capital costs
associated with fuel component controls implemented solely to
achieve such emissions reductions were amortized only over that
period.  In the NPRM, EPA used a 4.5 month amortization period.
Members of the refining industry and others commented that a 4.5
month period is too short because the high ozone period is 4.5
months long at terminals and refiners will have to begin making
RFC in advance to ensure that the fuel distribution system up to
and including terminals will have turned over before the starting
date.  Previous analyses performed in support of EPA's RVP
control program showed that fuel had to be produced between 0 and
60 days prior to arrival at a terminal.  Thus, EPA has extended
the RFC production period to 5.5 months, or 168 days.

     The amortization of capital costs estimated by B & M over
the high ozone season was accomplished by the following steps.  B
& M evaluated the cost of reformulating gasoline by running both
winter and summer refinery models, which operated off the same
set of capital equipment.  Thus, the seasonal differences in fuel
quality and production volumes were analyzed by the model and any
capital equipment built to meet one season's requirements (in
this case the summer's) was available during the other season as
well.  B & M's total annualized capital cost  (in $ per year)  for
each incremental fuel scenario was first multiplied by the ratio
of the capital recovery factors described above and then divided
by the volume of gasoline production for the ozone period.
Incremental annual operating costs (again in $ per year)  were
taken from both the winter and summer periods and divided by
summer RFC production, though practically speaking, those
occurring in the winter were quite small and generally
negligible.  Thus, the operating cost per gallon RFC was
essentially that determined in the summer portion of the model.

     Since the studies by Turner, Mason, and Co. for the Auto-Oil
program and WSPA were completed solely on an annual basis, it was
necessary to adjust the costs reported in those studies by
amortizing the capital related costs over the high ozone season.
This was accomplished by separating the total manufacturing cost
into an operating cost and an annualized capital cost,
multiplying the capital cost component by the ratio of 12 months
to 5.5 months and by the ratio of the capital cost recovery
factors described above, and then adding the adjusted capital
cost to the operating costs to obtain the adjusted manufacturing
cost.  This adjustment made their costs consistent with the
methodology used for the B & M data.

c.   Consideration of Regional Costs

     The refinery models used to estimate the costs of
reformulating gasoline were designed to model refinery operation
changes on a regional basis.  Consistent with this approach,  EPA

                               321

-------
directed B & M to divide the nation's refining capacity into
three regions:  Region 1 corresponds to PADD 1 (Petroleum
Administration for Defense District 1), Region 3 to PADD 3,  and
Region 2, which covers the remainder of the U.S.,  excluding
California.95  The modelling and the subsequent cost analyses
were based on reformulated gasoline production from these three
areas.

     EPA had B & M evaluate the refining costs for two separate
RFC demand scenarios which bracketed the possible opt-in
scenarios envisioned for the RFC program.  In one, named scenario
"A," the covered areas included the mandated nine extreme and
severe areas  (minus those of California), and those areas which
had opted-in to the RFG program as of the summer of 1992.  The
second,  named scenario "B," included the nine mandated areas
(minus those in California),  all ozone nonattainment areas which
have currently opted into the RFG program (prior to Kentucky's
recent decision),  the Northeast states which are members of the
Ozone Transport Commission (OTC),  and all other serious and
moderate nonattainment areas (California areas excepted).  In the
case of the OTC states, the attainment areas were included as
well as the nonattainment areas.  This would mean that all the
gasoline sold in the OTC areas would be RFG.

     In anticipation of many more nonattainment areas opting-in
to the RFG program, scenario "B" was chosen and analyzed for the
cost of the RFG program.  At the end of the study, however,
additional areas and the OTC had still not opted into the RFG
program as anticipated by using the "B" scenario.   In fact,  very
few areas have opted into the program since the proposal was
made; the nonattainment areas of Kentucky are the only new
additions.  Analyzing the level of opt-in at this time, only 31
percent of the nation's gasoline (excluding California) would be
RFG, which is 22 percent less than that used in the study.  If
the level of opt-in remains at the current level,  it is likely
that the costs of the program will have been overstated.
However, if many more areas opt-in to the program and the level
of RFG approaches that which was modelled, then the costs will
more closely represent that of the case modelled.
     95California  is  imposing its own Phase II reformulated
gasoline requirements on gasoline  sold in California beginning
in 1996.  The California Phase  II  requirements are likely to
require more severe  reformulation  than that  of the Federal
Phase II reformulated gasoline  and to exceed the federal
performance standards.  Thus, it is likely that the Federal
Phase II standards will not  affect fuel composition in the
State of California.

                               322

-------
     The amount of RFC required was calculated by applying the
fraction of the total state population living in the covered
areas to the total state gasoline demand of the year 1990.  These
fractions were then expressed as a fraction of national gasoline
demand and listed in the draft Regulatory Impact Analysis.  The
calculated nationwide percentage of RFC was next adjusted upward
by 10 percent to account for "spillover."  Spillover is the
amount of RFC which is expected to be sold either outside a
covered area, or before or after the required period of the
program.

     EPA determined this spillover rate based on information
available on the oxygenated fuels program, on its own analysis,
and on input from B & M.  The oxy-fuels program has been
estimated to have a 5 percent spillover rate.  This program,
however, involves minimal reformulation and,  in some cases, the
fuels can even be blended at the terminals rather than at the
refinery.  The greater reformulation required to make RFC
requires that refiners begin producing it sooner to ensure that
RFC formulation will meet the emission reduction standards at the
furthest point from the refinery by the program start date each
year.  However, RFC costs are expected to be greater than those
for the oxy-fuel program.  Thus, there will be a greater
incentive to minimize unnecessary costs,  such as spillover.
Based on the difference between the two programs, EPA recognized
that the spillover rate for the RFC program would be larger than
that of the oxygenate program.  B & M, based on their expertise
in fuel distribution, recommended a 10 percent rate, which seemed
reasonable considering the differences between the two programs.
Hence, EPA decided to use 10 percent as the spillover rate.

     The national RFC production estimate based on the scenario
described above, was then subdivided among the three gasoline
producing regions, facilitating the modelling on a regional
basis.  To do this, B & M first placed each RFC area into one of
the three regions to determine regional demand for RFC.  B & M
then used the historical movement of finished gasoline to
estimate which regions would supply each region's fuel
consumption. Based on this approach, the projected amount of RFC
produced by each region is listed in Table VI-1.  The last column
of Table VI-1 lists the fraction of gasoline produced within that
region which is projected to be RFC.
                               323

-------
                           Table VI-1:
      Reformulated  Gasoline Production Among Various  Regions
Bonner & Moore Region
(PADDs Included)
Region 1 (PADD 1)
Region 2 (PADDs 2, 4,
and 5; excl . CA)
Region 3 (PADD 3)
Fraction of
Total RFC
Produced in
U.S.
0.21
0.24
0.55
Fraction of
Gasoline in
Region that is
RFC
0.96
0.36
0.55
     These figures are useful in understanding the cost data in
several ways.  The regional RFC production fractions listed above
were used to weight the costs for each region and develop a
national cost estimate.  The larger the contribution by any one
region to the quantity of RFC produced nationally, the larger the
effect of that region's costs on the overall RFC program cost.
Hence, the fact that most RFC will be produced in Region 3 is
useful in understanding the various trends in costs.  Also, the
fraction of gasoline produced as RFC in any one region is useful
in understanding why costs may be high in one region and low in
another.  The higher the percent RFC, the higher the relative
cost, since more severe reformulation of a refinery's gasoline
pool is likely required.  Since the assumptions regarding the use
of reformulated gasoline by the OTC are significantly higher than
what is now expected,  the costs estimated for Region 1 are higher
than those likely to occur.  The Region 1 costs would be
considerably lower if a larger non-RFG market had been assumed,
since refiners would have had more flexibility in producing their
entire product slates.

     For similar reasons, marginal costs to produce RFC may be
higher for small refiners than those estimated by the regional
refinery models, unless a sizeable market exists for conventional
fuel.  Given the apparent unlikelihood that the Northeast OTC
will use Title I of the Clean Air Act to require RFG in
attainment areas as well as non-attainment areas, it appears that
all regions of the U.S., even Region 1, will have sizeable
conventional gasoline demands.  This demand will provide a
profitable conventional fuel outlet for the fuel from the small,
high RFG cost refiners.

     In contrast to the Bonner and Moore studies, the above-
referenced Turner, Mason, and Co. studies performed for both WSPA
and Auto-Oil (and some of the studies performed for API)  assumed
all the gasoline produced in a specific PADD was reformulated.
                               324

-------
Because of the lack of refining flexibility expected from 100%
production of RFC, the costs derived from these studies are
likely higher than what would be experienced by the average
refiner under the expected level of participation in the program.
Hence, for the fuel parameter changes for which the Turner/Mason
costs had to be used, the overall cost and cost-effectiveness are
expected to be overstated and therefore somewhat conservative.

     There are a few cases when multi-regional costs were not
available for developing national costs.  This occurred in two
cases when the B & M refinery model data was supplanted by other
data determined to be more appropriate.  However,  to be
consistent with the multi-regional approach, it was necessary to
develop adjustment factors that would approximate a broader
source of costs.  These factors would adjust a single regional
cost to a (non-California) national cost.

     For these cases, the regional to national cost adjustment
was made by the same methodology used in the draft RIA.  In that
case, Turner/Mason, working for Auto-Oil,96 determined the
refining costs to reduce olefins, T-90 and aromatics, and to add
oxygen, in each PADD.  The cost from each PADD was integrated
with the projected production of RFC to develop regional
adjustment factors.  The following table summarizes the
adjustment factors used to adjust costs from certain regions of
interest to a non-California national cost,  representing the RFG
costs in all three regions.


                           Table  VI-2:
      Factors  for  Adjusting Regional  Costs to  National  Costs
Adjustment Made
PADD 5 to National
PADD 3 to National
PADD 1 to National
Cost
Cost
Cost
Adjustment Factor
1.38
1.14
0.89
     One comment on the methodology used in the proposal
presented results of an informal survey of several members of the
refining industry, and used these results to argue that EPA's
analysis understated the costs of Phase II RFG.  Although the
commenter acknowledged that the results were based on only a
fraction of the refiners who will be affected by this rulemaking,
the commenter summarized the results as follows.  First, a number
     96Turner, Mason  &  Co. ,  April 1992.

                               325

-------
of large refineries will not make RFC (i.e.,  they will only
produce conventional gasoline).   Second,  for the refiners that do
make RFC,  RFC will constitute a large fraction of the products
from that refinery.  Finally, costs will likely vary between
refiners,  and most refiners will be limited by one or two
parameters in making RFC.  These results led to the conclusion
that RFC costs will be higher than those proposed by EPA.

     While the commenter's interpretation that the survey results
indicates that costs will be higher than those estimated by EPA,
the survey can also be interpreted to indicate that costs will be
lower than those represented by the oil industry and thus support
EPA's approach to estimating costs.  First, the fact that not all
major refiners plan to produce reform argues for reduced costs,
not increased costs.  All the refinery models employed to
estimate RFC costs model all the refineries in a given region or
PADD.  These refineries include those whose process configuration
is advantageous for producing RFC and those whose processes are
not.  It is most likely that the refiners who will choose not to
make RFC are the ones who cannot do so in a cost-effective
manner.  If the production costs for such refineries were removed
from the calculation of regional costs,  the regional costs would
be lower than the averages calculated by the existing models.

     Second, how the connection between fewer refineries planning
to produce RFC and higher RFC fractions for those refiners should
affect the use of EPA's RFC cost estimates is not clear.  The oil
industry's refinery modelling assumed high RFC fractions for all
refineries and that is clearly more onerous than the actual
situation is likely to be.  EPA, through its refinery modelling
by Bonner and Moore, assumed more realistic RFC fractions and
allowed the refinery model to determine how to produce RFC.  One
industry comment on EPA's approach has been that it allows the
model to build too few, but very large,  units to satisfy the RFC
demand for an entire region or PADD.  This approach is actually
confirmed by the oil industry survey.  While EPA's approach could
potentially underestimate costs by allowing the refinery model
too much flexibility in shifting low and high emission components
around within this massive regional refinery, EPA finds no
evidence that this actually occurred.  When the model
occasionally produced very low costs for Region 2, which had the
lowest RFC fraction, EPA decided not to use these results, but
instead based its RFC costs on those estimated by Bonner & Moore
for Regions 1 and 3.

     Third, the survey indication that most refiners' production
of RFC will be limited by one or two fuel parameters is
consistent with EPA's modelling approach of focusing on
improvements in a single fuel parameter while holding all others
constant or allowing them to improve (i.e., lower emissions).
While the sophistication of imbedding the complex model into a
refinery model is still technically preferred, the survey results

                               326

-------
indicate that the added sophistication may not significantly
affect the results.

3.    Fuel Component Control Costs

     The costs of modifying the various fuel parameters have been
significantly updated relative to those in the draft RIA.  In the
draft RIA, it was generally necessary to extrapolate costs
representing only a single region or PADD to the entire nation.
Most often the region for which costs were available (i.e.,
California)  would not even be producing Federal Phase II RFG
since the California Phase II RFG standards are more stringent.
Although an attempt was made to develop national average costs
from that data,  such costs are almost inherently less accurate
then those developed by actually modelling the refineries of each
covered region.

     Since the time the draft RIA was developed, a number of
studies (performed by both B & M and Turner/Mason)  have evaluated
RFG costs in more than one region and generally provide much
improved estimates of fuel reformulation costs.  In addition, EPA
funded its own study.  The greater access to cost data allows EPA
more capabilities in determining and comparing RFG program costs.

     The following sections describe the costs estimated for
controlling each fuel parameter, including any adjustments made
which have not already been described.  After all fuel parameter
costs have been described and combined with the complex model to
estimate the cost-effectiveness of emission control, the results
will also be compared to those of refinery studies which imbedded
the complex model into the refinery model to generate cost-
effectiveness estimates directly.  The final costs used in this
analysis are summarized in subsection 5 below.   The derivation
of the final costs are shown in Tables VI-Al, VI-A2, VI-A3 and
VI-A4 in the Appendix.

a.    RVP Control Costs

     The B & M refinery model were used to project the cost of
RVP control.97  The cost of RVP control was modelled from 8.0 psi
down to 7.3 psi for the first increment,  down to 7.1 psi for the
second increment, and down to 6.5 psi for the third increment.
Although the ranges over which the parameter values changed for
each region did not agree exactly, the costs for the most similar
ranges were grouped together.  The fuel economy adjusted costs
for a RVP reduction from 8.0 to 6.5 psi were about 0.7C/gal.  As
shown below in section F,  reducing RVP to control VOC emissions
     97Lester Wyborny,  U.S.  EPA/FSSB,  "DOE and API Phase II Cost
Estimates," Memorandum to the Air  Docket,  November 4,  1993.

                               327

-------
is very cost effective all the way down to 6.5 RVP.   Although a
further reduction below 6.5 psi would likely be cost effective,
such reductions were not considered because,  as discussed in
later sections, it is not known whether driveability problems
would result at lower RVPs.  Also, no exhaust emission data exist
below 6.5 RVP, so possible increases in exhaust emissions could
not be ruled out.

     The low costs for RVP control estimated by the B & M model
cost data are corroborated by other studies which also have
evaluated the cost of RVP reductions.  The NPC study showed that
when the fuel economy effect is taken into account,  initial RVP
reductions down to 7.2 psi were virtually at no cost, while
further RVP reductions down to 6.5 psi were made at a 1.2C/gal
cost.  Although cost figures were not available from the DOE
study, DOE concluded that RVP costs for VOC control were much
cheaper and cost-effective relative to other any parameter
changes which would also reduce VOC emissions.

b.   Sulfur Control Costs

     The B & M regional refinery models were also used to
estimate the costs of sulfur control.  Sulfur control was
modelled down to 250 ppm for the first increment, to 160 ppm for
the second increment,  to 100 ppm for the third increment, and to
50 ppm for the last increment.  The sulfur costs for the value of
138 ppm sulfur used in setting the NOx standard was interpolated
from the other data using a curve fitting program available in
Cricket Graph.  As shown in Table VI-A2 located in the appendix,
the B & M Region 2 costs are quite low compared to those for
Regions 1 and 3.   This is likely due to the low fraction of RFG
being produced in Region 2, as described above.  These very low
costs could be realistic, representing the ability of refineries
to shift blendstocks between conventional fuel and RFG to reduce
sulfur very inexpensively.  However, they could also represent an
unrealistic flexibility to do this, given that the entire
region's refinery capability was combined into a single refinery.
Insufficient time prevented further analysis to resolve this
issue.  In order to avoid the possibility of underestimating the
cost of sulfur control, the Region 2 costs were excluded from the
analysis and the nationwide cost was generated by weighting the
Region 1 and 3 costs using renormalized weightings from Table VI-
1 above.

     The B & M Region 3 sulfur control costs were compared to
Region 3 (PADD 3)  cost data generated by Auto-Oil and API,  both
studies performed by Turner, Mason, and Co.  These other studies
were adjusted to a 10% rate of return on investment and the
capital costs were amortized over the high ozone season to make
the studies consistent.  As shown in Figure VI-A1 (see the
Appendix),  the Auto-Oil costs turned out to be very similar to
the B & M sulfur costs over the range from 340 ppm down to 100

                               328

-------
ppm; about 1.5-1.6C/gal.   The API costs (based on refinery
modelling cases VAT and VH),  however,  are about twice that of the
Auto-Oil and B & M costs over a smaller range.  The API parameter
data for those refinery model runs indicated that there were also
some increases in E200 and E300 along with the sulfur reductions.

     It is feasible that the refinery model used by API, with the
complex model integrated in the model, would have made these
changes to realize the necessary reductions in VOC.   However, it
is doubtful that the model would have actively reduced these
parameters considering their higher cost per pollutant reduced as
determined in this study.  Instead, it is likely that they
increased as a result of refinery changes associated with
reductions in sulfur or other parameters.   If one assumes that
E200 and E300 were purposefully reduced by the API version of the
model,  thereby incurring a reformulation cost, the cost for that
refinery model run can be adjusted to account for the full E200
and E300 reduction.  The E200 and E300 costs derived in this
analysis were used for that purpose.  The adjusted API costs are
indicated on Figure VI-A1 as a dashed line.  Even after this
adjustment, the API costs are still substantially higher than
that those from B & M and Auto-Oil.

     Because of the importance that sulfur reductions play in
controlling NOx emissions, EPA held a number of discussions with
DOE and API in an attempt to determine the differences between
the API costs and those generated by B & M and other studies,
like the Auto-Oil study compared here.  The discussions centered
around several documents which summarized the approaches or
critiqued approaches of these models.   These documents include a
list of the underlying assumptions used in the API refinery
modeling runs by Turner/Mason, a critique of the B & M refinery
runs by Turner/Mason, and a reply by B & M to the Turner/Mason
critique.98  The numbers used  in this  final analysis represent
EPA's best estimate of the costs of sulfur control,  taking all of
these other issues into account.

c.   Olefins Control Costs

     EPA based its cost of olefin reductions on those estimated
by Turner/Mason for WSPA99 instead of  those generated by B & M.
The B & M estimates were not used because much of the olefin
content was used to make oxygenate  (in the form of TAME or MTBE)
prior to the olefin reduction refinery modelling runs.  Since the
     98"Documents  Concerning Refinery Modeling for the
Reformulated Gasoline Program,"  EPA Memorandum From Lester
Wyborny, RDSD to  the Docket,  .

     "Turner, Mason  & Company,  November 18,  1993.

                               329

-------
olefin level was already partially reduced,  the cost of reducing
olefins in a subsequent step may not reflect the true cost of
olefin control at lower oxygen levels.  Based on the results of
the cost-effectiveness analyses described below, EPA does not
expect the overall RFC pool oxygen content to increase much above
2.1 weight percent prior to the reduction of olefins for NOx
control,  which should reduce the olefin demand from the levels
predicted by the refinery modelling.

     Turner/Mason estimates of olefin control costs for both WSPA
and Auto-Oil were considered for use here.  The WSPA estimates
were determined to be more desirable for cost-effectiveness
purposes because they were expressed as a series of reductions
instead of one single large reduction.  To compensate for the
fact that the WSPA study was based solely on California
refineries,  the WSPA olefin costs were increased by the 38
percent adjustment factor described above.  The WSPA analysis
contained cost data for olefin reduction in increments of 11 down
to 8 percent, and then down to 5 percent.  The cost to decrease
olefins from 11 to 8 percent was estimated to cost about
0.4C/gal, while further reductions to 5 percent were projected to
cost 2C/gal.

d.   Oxygen Control Costs

     For the February 26, 1993 proposed rule, EPA based its
oxygenate costs on the Auto-Oil (A/0) study of the RFG program,
which relied upon MTBE as the oxygenate.  The purchased price of
MTBE was estimated by A/0 to be about 60C/gal.  In comments
received on the proposed standards, it was suggested that the
analysis in the draft RIA underestimated the cost of oxygenate
because it used current contract prices for MTBE rather than
projections of future prices.  As will become evident below, the
final rule is based on much higher oxygenate market prices, and
even includes a price range of oxygenate costs to account for
potential future price fluctuations.  Furthermore, the choices of
oxygenates was expanded beyond MTBE to include TAME and ethanol.

     For the final rule,  the per-gallon costs for an oxygen
increase from 2.1 percent (the control level under Phase I
averaging) to 2.7 percent were initially taken from the B & M
refinery model cost data.  The model could choose among the
production or purchase of TAME and MTBE, or the purchase of
ethanol.   TAME and MTBE were generally produced at the refinery
although MTBE was sometimes purchased from the outside, while the
ethanol was always purchased from the outside.  Based on this
approach and the built-in economic criteria, the model generally
preferred ethanol as the most cost effective choice for the
addition of oxygen for Phase II RFG.

     Table VI-3 below summarizes the oxygenate prices used by the
model.  To account for the historical fluctuations in oxygenate

                               330

-------
prices, the cost analysis was expanded to consider a range of
oxygenate prices.  Table VI-3 also contains the ranges in
oxygenate prices which were used in this expanded analysis.  The
Phase I cost analysis located in section V contains a description
of the reasoning behind the values used in the oxygenate price
range.  In general, the range represents the range in oxygenate
prices observed over the last several years.
                   Table VI-3:   Oxygenate  Costs
Oxygenate
Ethanol
MTBE
Methanol
B & M Model
Costs
$1.20
$1.05
$0.57
Cost Range Used
Before Subsidy $1.20
After Subsidy $0.66
by EPA
- $1.60
- $1.06
$0.80 - $1.05
$0.40 - $0.57
     The Phase II RFG oxygen addition costs for Regions 1, 2 and
3 determined by the B & M refinery model were then adjusted to
reflect this revised range of possible oxygenate prices.  In
Table VI-4 below, the cost of oxygenate addition determined by
the B & M refinery model is listed first followed by the adjusted
costs.  The regional oxygenate costs were combined into a U.S.
cost by weighting each regional cost by the fraction of RFG
produced by that region (see Table VI-1).   The national average
cost was used to determine the cost-effectiveness of adding
oxygen to Phase II RFG (above the level of 2.1 wt% oxygen).
          Table VI-4:  Phase  II  RFG  Oxygen Addition  Costs
Geographic Area
Region 1
Region 2
Region 3
U. S.
Estimated RFG Costs
B & M Refinery Model
«Vqal RFG)
1.37
1.12
1.16
1.20
RFG Cost Range
«Vqal RFG)
0.62 - 1.23
(0.25) - 0.76
0.32 - 1.00
0.24 - 0.99
(0.61 average)
                               331

-------
e.   E200 Control Costs

     E200 control costs were developed using EPA's in-house
version of the B & M model.  No other studies evaluated the costs
of controlling E200 alone.  (E200 was monitored and varied in a
number of the refinery studies which imbedded the complex model
in a refinery model and which are evaluated in section G below.)
While these costs were not reviewed by B & M personnel, in
general, the E200 costs seemed reasonable for the changes in
gasoline characteristics that occurred.  However, because the
refinery model was not specifically designed to evaluate all of
the E200 control technologies currently available, the costs may
be somewhat high.  The B & M model estimated the costs of
controlling E200 in 3 percent increments.

f.   E300 Control Costs

     E300 control costs were taken from the Auto-Oil study.
Attempts were made to derive the E300 costs in-house using the B
& M model.  Results of these modelling runs were insufficient,
however, because the costs quickly escalated for even small
increases in E300.  A review of the refinery processes used by
the model to increase E300 revealed that the model was not
programmed with sufficient analytical tools to make accurate
estimates.  Thus, the Auto-Oil E300 costs were used instead.
Although WSPA costs were also available, those costs were from
refinery runs of Padd 5 refiners, which are not even covered by
the Federal RFG program.  The Auto-Oil costs from Padd 3 are
therefore considered more relevant to this study.  Furthermore,
because both studies were done by Turner/Mason, the costs should
have been be derived using nearly identical approaches.

     The E300 costs based on Auto-Oil were calculated from T90
control using a formula derived by EPA to convert to E300, the
derivation of which is described in section III.100  The costs
were adjusted upward by 14 percent to project PADD 3 costs to
nationwide costs.  For the E300 study for Auto-Oil, Turner/Mason
used increments from 84 to 88% evaporated, 88 to 91% evaporated,
and 91 to 96% evaporated.  The adjusted cost to increase E300 to
88 percent cost 1.10/gal, to increase to 91 percent cost
1.9C/gal, and to increase to 96 percent cost about 4C/gal.

g.   Aromatics Control Costs

     Aromatics control costs were derived from the B & M refinery
model by EPA.  The aromatics concentration was reduced to 28
     100E300 was calculated from T-90 using  the  formula:  E300 =
155.033 -  (T90 x 0.22)

                               332

-------
volume percent for the first increment,  down to 24 percent for
the second increment,  and down to 20 percent for the third
increment.  When the Region 2 costs were compared to Regions 1
and 3, it was decided that their costs should not be not used
because of their much lower values.  As stated above,  because
there was insufficient time to verify that the less expensive
parameter reductions generated by the Region 2 refinery model for
some parameters were made in a realistic fashion, EPA elected to
err on the conservative side and drop the costs from the cost
analysis.  The aromatics costs were therefore based only on
Regions 1 and 3.

     The B & M aromatics costs were compared to those generated
by WSPA and Auto-Oil (which, as PADD 5 models, were adjusted to a
nationwide basis by the 1.38 factor developed previously).   The
WSPA costs were determined to be 2.9 cents per gallon over a 3
percent aromatics change between 25 and 22 percent.  The Auto-Oil
costs proved to be 12 cents per gallon over an aromatics
reduction from 45 vol% to 20 vol%.  To provide for a comparison
among the data, the Auto-Oil cost was ratioed down to represent a
3 percent aromatics change and the cost was determined to be 1.7
cents per gallon.  Both the WSPA and the adjusted Auto-Oil costs
are higher than that generated from the B & M models which, for a
reduction from 25 to 22 percent, was projected to cost 1.3 cents
per gallon.  The methodology used by the B & M model is different
from that observed in the Auto-Oil study and results in less
aromatics reduction upon the addition of oxygen.  Considering
that aromatics was not a cost-effective means for reducing any of
the controlled pollutants even at the lower B & M control cost,
and due to time and resource restraints, EPA did not evaluate
further the reason for the cost differences among the different
models.

     The first incremental aromatics reduction from the B & M
cost data was not needed in the cost-effectiveness analysis
because the aromatics concentration in the Phase I RFC parameter
data, which formed a starting point for the Phase II analysis,
had already decreased to a level below the lower value of that
range.  As described more fully below in the parameter
interrelationships section, it was necessary to adjust the Phase
I parameter values to ensure that the final Phase I RFC parameter
values would reflect a sufficient aromatics reduction to meet the
toxics requirements.

4.    Fuel Economy Effects

     Reformulating gasoline to reduce emissions can affect the
vehicular fuel economy.  In the draft RIA, EPA found that the
change in a fuel's heat content is currently the best indicator
of the change in fuel economy between fuels of different
compositions.  In this section, the methodology for estimating
the costs attributed to the changes in the heat content of RFC

                               333

-------
will be described.  The changes in fuel economy are associated
with specific RFC parameter changes,  including increased oxygen
content, reduced RVP, reduced aromatics content,  and increased
E300.  As found in the draft RIA,  most other fuel components are
expected to have relatively minor effects on fuel economy and
thus the impacts of these effects were not included in the
analysis.

a.   Changes in Fuel Economy

     In the draft RIA, EPA listed the estimated fuel economy
impacts of various fuel parameter changes as determined by Auto-
Oil and ATL (under contract with EPA) testing.  However, the test
data for some of the parameter changes varied significantly in
magnitude.  This variance is largely attributed to measure
uncertainty in both the fuel and emissions,  as well as the effect
of changes in other fuel parameters that were not controlled by
the test programs and variability in vehicle operation between
tests.  In the draft RIA, EPA found that fuel economy generally
changed in proportion to fuel heat content.   No comments that
were accompanied by any data or engineering analysis not already
considered in the draft RIA were received challenging this
finding.

     In order to estimate the effect of various fuel
reformulations on heat content and, therefore, fuel economy, EPA
had B & M calculate the heat content of reformulated gasoline as
they ran their refinery models for economic purposes.  This was
done for oxygenate addition and RVP reduction.  These calculated
effects were used in lieu of the test data,  since the test fuels
did not necessarily reflect the other changes in gasoline
properties likely to accompany either oxygen addition or RVP
control.  One drawback of B & M calculating the fuel economy
effect for RVP is that the evaporative effects were not
considered; the RVP adjusted costs for fuel economy effects are
expected to be somewhat overstated for this reason.  B & M did
not evaluate the fuel economy impacts of adjusting E300 or
aromatics, and thus the available test data was used.  Since the
test data also showed a very minor fuel economy impact for olefin
reductions, B & M was not asked to evaluate olefins for their
fuel economy effect and any effects were considered negligible.
Sulfur has not been found to cause a fuel economy impact and also
was not studied. The fuel economy impact estimates made from the
test data and B & M study are listed below in Table VI-5.
                               334

-------
    Table  VI-5:  Fuel Economy Impacts of Fuel Parameter Control
Parameter
Studied
Oxygen - MTBE



Oxygen - EtOH

Oxygen - ETBE
RVP


T90 (E300) *

Average :
T90/E300
Aromatic s
Study
Auto-Oil I
Auto-Oil VI
EPA ATL (Phase I)
Bonner and Moore
Auto-Oil VI
Bonner and Moore
Auto-Oil VI
Auto-Oil VI
EPA ATL
Bonner and Moore
Auto-Oil I
EPA ATL

Auto-Oil I
Energy Impact per Unit
Parameter Change
-0.9/wt% increase
-0.70/wt% increase
-0.5/wt% increase
-1.05/wt% increase
-0.61/wt% increase
-0.95/wt% increase
-0.64/wt% increase
2.6/psi decrease
0.7/psi decrease
0.3/psi decrease
-0.02/deg F increase
(-0.0732/%E300
decrease)
0.003/deg F increase
(0.013/%E300 decrease)
-0.009/deg F increase
(-.03/%E300 decrease)
-0.12/vol %
*  E300 was calculated from T-90 using the formula: E300 = 155.033 -  (T90 x
0.22)
     For adding oxygenate, B & M  first  calculated  that
conventional gasoline contains about 4.75 million  BTU per barrel,
and MTBE and ethanol, the oxygenates blended  in  the  largest
quantities in their analysis, contain 3.92  and 3.19  million BTU
per barrel, respectively.  They then calculated  the  decrease in
energy that occurred by supplanting some of the  conventional
gasoline with oxygenate to yield  a fuel containing 2.1  weight
percent oxygen  (an 11.5 volume percent  addition  of MTBE or 6.0
volume percent addition of ethanol).  The calculation showed that
the heating value of the resulting gasoline-MTBE blend  would be
2.1 percent less energy dense than conventional  gasoline,  or that
the gasoline/ethanol blend would  be 1.9 percent  less energy dense
than conventional gasoline.  As an approximation,  EPA assumed
                               335

-------
that either oxygenate would reduce the energy content of gasoline
by about 2 percent.

     B & M made a similar calculation for reducing RVP.   In their
analysis, butane was the only component removed to control RVP;
an eight percent decrease in butane content resulted in a 1.2
percent increase in heating value.  EPA then calculated that
based on an RVP blending value for butane of 60 psi, each two
percent reduction in gasoline butane content would result in
about one psi change in RVP.  The application of these two
figures determined that each psi decrease in RVP would result in
about a 0.3 percent increase in fuel economy.

b.   Fuel Consumption Costs

     The percentage changes in fuel economy identified above can
be converted to a monetary cost by multiplying them by the retail
(less taxes) price of gasoline, here estimated to be 85 cents per
gallon.  This price is 5 cents per gallon higher than that used
in the draft RIA to accommodate the cost of producing Phase I RFG
(the high end of the range presented in Section V).   This fuel
economy cost is then added to the refining cost.  The effects of
this calculation are demonstrated in the following summary table.

5.   Summary of Fuel Parameter Control Costs

     The refinery costs determined by the methodologies explained
above are listed below in Table VI-6.  The costs are expressed in
cents per gallon per unit change in the given parameter (i.e.,
for RVP, the unit quantity is one psi) as indicated by the last
column.  In all cases, the cost is applicable to the range
incremental to the previous range.  For example, the cost for
sulfur reduction down to 160 ppm covers the cost for reducing
sulfur from 250 ppm to 160 ppm, and does not include the cost of
reducing sulfur to 250 ppm.  To determine the cost for reducing a
parameter over part or all of a range, the change in fuel
parameter value (final value minus the starting value)  would be
multiplied by the fuel economy adjusted cost.

     The first three columns of the summary table show the cost
in each region, as determined by each individual regional
refinery model.  Where a space for a regional cost is left blank,
that cost was either not determined by the B & M refinery model,
or the cost was not used for the reasons described above.   The
regional costs were factored into a national cost using the
expected fraction of RFG produced by each region, as described
above.  The national cost is then adjusted by the fuel economy
effect calculated for that parameter, one exists.
                               336

-------
Table VI-6: Incremental Costs for Individual Parameter Changes
Fuel
Parameter &
Control
Level
RVP to 7 . 3
RVP
7.1 to 7.3
RVP
7.3 to 6.5
Sulfur to
250
Sulfur
250 to 160
Sulfur
160 to 100
Olefins to
8.0
Olefins
8.0 to 5.0
Oxygen
2.1 to 2.7
Aromatics to
28
Aromatics
28 to 24
Aromatics
24 to 20
E300 (+4)
84 to 88
E300 (+3)
88 to 91
E300 (+5)
91 to 96
E200 (+2, +3)
to 58
E200 (+3)
58 to 61
E200 (+3)
61 to 64
National
Average
Cost
0.412
0.416
0.427
0.00203
0.00625
0.0125
0.179
0. 922
1.14
0.0664
0.305
0.367
0.285
0.645
0.634
0.125
0.439
0.989
Fuel
Economy
Effect
(c/gal/x)
0.220
0.220
0.220
-
-
-
-
-
0.85
0.085
0.085
0.085
0.026
0.026
0.026
-
-
-
Fuel
Econ.
Adjusted
Cost
0.192
0.196
0.207
0.00203
0.00625
0.0125
0.179
0. 922
1.99
0.152
0.391
0.452
0.311
0.671
0.660
0.125
0.439
0.989
Units
(for columns
2 and 5)
(c/gal/x)
x = psi


x = ppm


x = %

x = %
x = %


x = %


x = %


                              337

-------
D.    Derivation of Fuel Parameter Interrelationships

     Controlling one parameter of a fuel (a fuel property or
component) can affect the levels of other fuel parameters, since
the refiner will compensate for the change by adjusting the fuel
composition to meet minimum gasoline specifications in the manner
which is most cost effective for its refinery.  This effect of
one fuel parameter on another is hereafter referred to as an
interrelationship.  To be able to determine the emissions that
would result from a change in a single fuel parameter, it is
necessary to estimate the final fuel composition based on these
interrelationships.

1.    Methodology

     For the draft RIA, factors representing fuel parameter
interrelationships (or fuel component interactions, as they were
called in that document)  were developed from the available
refinery modelling data.   In response to this information, EPA
received several comments.  One stated that the change in fuel
properties was underestimated for some parameters and
overestimated for others.  Specifically, the commmenter claimed
that EPA appeared to recognize the dilution effects of adding
oxygen,  but that it was not clear that this was always included
in the analysis.  Since specific references to the areas where
EPA overlooked the impact of adding oxygen, or other cases where
EPA over- or underestimated the impacts of various fuel
parameters, it is difficult to respond directly to this comment.
As explained below,  the analysis of parameter interrelationships
has been revised based on new information and is believed to be
reasonable.  When an interrelationship between two parameters was
identified, this was used consistently throughout the analysis.

     A second comment raised a concern with the fact that EPA
used data from a single PADD when developing these parameter
relationships, assuming that the effects seen in this one PADD
would be experienced in other PADDs (which is not necessarily the
case).   The refinery modelling data presented in the draft RIA
and used to develop parameter interrelationships was limited in
its applicability to the RFC program for several reasons.  The
data was based largely on a single PADD (usually PADD 5, in some
cases PADD 3) model,  not on multiple PADD data (which could be
averaged to develop a national factor).   The refinery costs and
operating parameters for one PADD are unlikely to be
representative of what a typical refinery elsewhere in the
country would experience.  Furthermore,  refineries in PADD 5,
which supply California's fuel, are somewhat different than
refineries elsewhere because of the greater percentage of
hydrocracker capacity in the PADD, and because of the heavier
crude feedstocks used there.  California is developing its own
RFG program and Federal Phase II RFG will not likely be produced
in PADD 5.  Hence, for most fuel parameter changes analyzed in

                               338

-------
this study, parameter interrelationships were developed from new
refinery modelling data for multiple regions, generated by the
Bonner and Moore (B & M)  refinery model; exceptions to this are
discussed below.

     Data was generated for each of the three regions defined by
B & M (discussed in subsection 2.c above).   The methodology for
determining parameter interrelationships was similar to that used
to estimate costs.   Each parameter interrelationship was
determined for the entire range over which the primary parameter
was changed.  The interrelationship was expressed as a ratio of
the percent change in the secondary (uncontrolled)  parameter to
the overall change in the primary  (controlled)  parameter.  For
example, if reducing the sulfur level from 250 ppm to 160 ppm
resulted in a change in olefin content from 5.5 vol% to 5.3 vol%,
then it was assumed that for every change of -90 ppm sulfur
(within the approximate range of 250 to 160 ppm),  the olefins
content would change -3.8% [=(5.5-5.3)/5.5 x 100].   The
interrelationship factor would thus be 0.042% olefins/ppm sulfur
[=-3.8%/-90ppm].

     Interrelationship factors were developed for each of the
three refining regions.  A single interrelationship factor
representing the average national impact was then developed by
weighting the factor for each region by the fraction of total
(national)  RFC produced by that region  (see Table VI-1 above).

     These factors were included in the RFC cost-effectiveness
analysis only if it was determined that the factors represented a
significant and consistent relationship between the two
parameters.  These engineering judgements were made on the basis
of the magnitude of the factor, the consistency of the trend
between the three regions, and the confidence in the refinery
modelling outputs (i.e.,  did the results agree with what would
likely happen in a refinery).

     The following table presents the results of this analysis of
parameter interrelationships.  The factors listed in the last
column are national parameter interrelationship figures derived
from the various refining regions and were used in the RFC cost-
effectiveness analysis to represent the interrelationship of
various fuel parameter changes.  Details of the derivation of
these factors are provided below.
                               339

-------
       Table VI-7:  Fuel  Parameter Interrelationship Factors
2 .

a.
Controlled
Parameter
Phase I :
Oxygen
Benzene
RVP
Sulfur
Olefins
RVP
Oxygen
Aromatics
E200
E300
Range of Control
Up to 2.1 wt%
Down to 0.95
vol%
Down to 8.0 psi
Down to 250 ppm
250 to 160 ppm
160 to 50 ppm
Down to 250 ppm
250 to 160 ppm
160 to 50 ppm
Down to 6 . 9 vol%
6.9 to 5.0 vol%
Down to 7.3 psi
7.3 to 6.9 psi
6.9 to 6.5 psi
2 to 2.7 vol%
2 to 2.7 vol%
Down to 28 vol%
28 to 24 vol%
24 to 20 vol%
Up to 54 %evap
54 to 57 %evap
57 to 60 %evap
Up to 88 %evap
88 to 91 %evap
91 to 96 %evap
Uncontrolled
Parameter
Sulfur
E200
E300
Aromatics
Olefins


Aromatics


E200

E200


Aromatics
E200
E300


E300


E200


(% Change in Uncontrolled)/
(Real Change in Controlled)
91% of Baseline Value17
114% of Baseline Value27
102% of Baseline Value
80% of Baseline Value
0 .033
0 .011
0 .046
0 .008
0 .013
0 .042
1 .791
1 .699
4 .146
2 .643
5 .663
-7 .297
1 .064
-0 .583
-0 .336
-0 .392
0 .221
0 .338
0 .340
1 .282
0 .600
0 .707
      represented as a straight percent change from the baseline value (see text).
      27 For VOC control region 1, the increase in E200 was only 111% from baseline for Phase
      I RFG, reflecting the fact that more light compounds are removed to reduce  the RVP to
      the lower level of 7.8 psi in this area.
Individual  Fuel  Parameter Interrelationships

Phase I RFG Parameters
      Although the  Phase  II RFG analysis  is  separate from that
for  Phase  I,  it  is important  to understand  the  parameter
                                   340

-------
interrelationships of the Phase I requirements  (2.1 wt% oxygen,
0.95 vol% benzene, and RVP at 8.0 psi) since they determine the
starting fuel composition of the Phase II fuel.  The standards
for Phase II RFC are based on the percent reduction in
emissions from the Clean Air Act baseline fuel emissions.
However, the cost-effectiveness of Phase II RFC is based solely
on the costs incremental to those incurred by refiners in
meeting the Phase I standards.

     The Bonner and Moore refinery modelling data obtained since
the draft RIA included a revised analysis of the production of
Phase I RFC.  This data was reviewed to determine the Phase I
parameter interrelationships.101  The methodology outlined above
for determining the B & M parameter interrelationships for each
parameter change could not be followed for the Phase I data
because the Phase I refinery model run included changes in all
three parameters.  To represent the fuel parameter changes
resulting from implementation of the Phase I RFC requirements,
the overall percent change from baseline for each uncontrolled
parameter affected by these changes (aromatics, olefins,
sulfur, E200, and E300)  was calculated from the B & M data.
The results of this calculation are presented in Table VI-8
below.
  Table VI-8:  Phase I RFG Interrelationships (Bonner & Moore]
Controlled
Parameter
Oxygen
Benzene
RVP
Range of Control
Up to 2.1 wt%
Down to 0.95
vol%
Down to 8 . 0 psi
Uncontrolled
Parameter
Sulfur
E200
E300
Aromatics
% of Original
(Baseline) Value
94
105
104
89
     The Phase I parameter interrelationships determined from
the B & M data were different than those presented by other
studies.  Specifically, the addition of oxygen appeared to have
little impact on the aromatics content beyond simple dilution,
even though other studies, such as that performed by Auto-Oil,
     101Bonner and Moore Management Science,  "Study of the
Effects of Fuel Parameter Changes on the Cost of Producing
Reformulated Gasoline," Prepared for EPA under contract through
Southwest Research Institute, (see "DOE and API Phase II Cost
Estimates," EPA Memorandum from Lester Wyborny, FSSB, to the
Air Docket, November 4, 1993.)
                               341

-------
show that aromatics would be reduced beyond the level achieved
by dilution since the oxygenates can replace octane usually
provided by aromatics.  The small change in aromatics by the B
& M refinery model is probably caused by the refinery model's
large use of olefins to produce oxygenates.  Since some of the
increase in octane from the addition of oxygenate was offset by
the loss in high octane olefins, aromatics could not decrease
as much as expected, resulting in higher aromatic values.
Because this approach is not expected to be taken by refiners
who also must meet the toxics requirement, a more average
approach than that by B & M was adopted.

     The data presented in the draft RIA,  based on Auto-Oil
studies, showed a slightly different relationship between the
various fuel parameters as a result of the Phase I RFC changes.
Although the Auto-Oil data was based on modelling of refineries
in a single PADD, most of these relationships were similar to
those seen in the new Bonner & Moore data.  However, the change
in aromatics as a result of the addition of oxygen was more
reasonable in the Auto-Oil data.  The relationships between
parameters calculated from the Auto-Oil data are shown in Table
VI-9.
     Table VI-9:  Phase I RFG Interrelationships (Auto-Oil]
Controlled
Parameter
Oxygen
Benzene
RVP
Range of Control
Up to 2.1 wt%
Down to 0.95
vol%
Down to 8 . 0 psi
Uncontrolled
Parameter
Sulfur
E200
E300
Aromatics
% of Original
(Baseline) Value
86
120
104
71
     The parameter interrelationships for the Phase I RFG used
as the starting point for the Phase II RFG cost-effectiveness
analysis were determined from a combination of Bonner and Moore
and Auto-Oil data, since neither study presented the best
relationships between parameters for Phase I RFG.  For each
uncontrolled parameter,  the percent change from baseline
calculated from these two studies was averaged for each region
for which data was available.  The national weighted average
percent change was then calculated and used in the cost-
effectiveness analysis.   These combined values are listed in
Table VI-7.
                               342

-------
b.   Impact of Reducing Sulfur Content on Other Fuel Parameters

     There are essentially three options available to reduce the
sulfur content of gasoline.  The majority of sulfur in finished
gasoline results from the gasoline components coming from the
FCC unit.  One option to reduce the sulfur level is to
desulfurize the feed to the FCC unit; this would not impact
other finished gasoline parameters.  The other options are to
desulfurize the output of the FCC unit or to use less FCC
gasoline in the finished product.  These options would result
in a reduction in the level of olefins in the final product,
since these streams are also high in olefin content.  This
reduction in olefins decreases the octane of the gasoline,
which must be made up by increasing the level of oxygen,
aromatics, or alkylates in the finished gasoline.

     The refinery modelling data used in this analysis indicated
that reducing the sulfur content of reformulated gasoline may
have some impact (depending on the refining region and the
level of sulfur control) on olefins, aromatics, E200, and/or
E300.  This data is presented in Table VI-10.  In addition, the
table shows the percent change in the uncontrolled parameter
resulting from each change in sulfur content.
                               343

-------
                    Table VI-10:
Data Used to Develop Sulfur Interrelationship Factors
Controlled
Parameter
Sulfur
Sulfur
Sulfur
Sulfur
Reqion
1
2iy
3
1
2iy
3
1
2iy
3
1
2iy
3
From
337 ppm
250
160
454
250
160
441
250
160
337
250
160
454
250
160
441
250
160
337
250
160
454
250
160
441
250
160
337
250
160
454
250
160
441
250
160
To
250 ppm
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
250
160
50
Uncontrolle
d Parameter
Olef ins








Aromatics








E200








E300








From
6.9 vol%
5.5
5.0
6. 9
6.9
5. 0
6.9
6. 9
5.0
29.7 vol%
29.7
27 .2
31.2
29.7
26.5
31 .1
30.5
28 .9
51.5 %evap
51 .4
51.1
53 .7
53.3
52 .3
56.5
54 .5
53 .5
85.8 %evap
85.5
85 .9
85.1
86 .4
86.5
85 .6
87.7
86 .7
To
5.5 vol%
5 .3
4 .1
6 .9
6.9
5 .0
6.9
6 .9
5.0
29.7 vol%
29.0
26.3
28.3
26.3
24.4
30.5
30.3
27.4
51.1
%evap
50.3
52.1
55.9
53.0
51.8
57.4
54 . 9
56.8
85.8
%evap
85.9
86.6
86.8
87.2
87.3
86.6
86.6
88.2
% Change
-20.3
-3 .6
-18.0
0.0
0.0
0.0
0.0
0 .0
0.0
0 .0
-2.4
-3 .3
-9.3
-11.4
-7.9
-1. 9
-0.7
-5.2
-0.7
-2 .1
2 .0
4 .1
-0.7
-1.0
1.6
0 .8
6.2
-0.1
0.5
0 .8
2.0
0.9
1.0
1 .1
-1.3
1 .7
                         344

-------
     1 Data from Region 2 was questionable (as discussed in the text)  and was excluded
     from the analysis.

     The data in Table VI-10 shows that only in  Region  1 does
reducing the sulfur content of the fuel have an  impact  on
olefins; in the other two Regions no impact is indicated.  This
could be a result of the fact that refiners in Region 1 had
little flexibility when reducing sulfur content,  since  96% of
the gasoline produced in that region was  reformulated  (under
the scenario analyzed here), and thus, refiners  were forced to
control sulfur in a manner which also affected olefins.  For
this region, it is likely that the sulfur was removed from the
output of the FCC, causing the removal of olefins at the same
time.  The Region 2 interrelationship numbers may be
questionable, because (as discussed previously)  the data also
indicated that there was less cost than expected associated
with controlling sulfur levels in finished RFG.   It appears
that the B & M model overestimated the existence of
hydrotreating capacity in this region, or overoptimized the
gasoline production options in this region since less than one-
half of the fuel is reformulated.  In contrast,  although there
also appears to be no impact on olefins of reducing sulfur for
the Region 3 data, there was a reasonable cost associated with
sulfur control in this region.  The lack  of an
interrelationship between sulfur and olefins may be explained
by the fact that the Region 3 refineries  may elect to
desulfurize the feed to the FCC unit without affecting  the
olefin level.

     Other studies, most notably the work performed for Auto-Oil
and referenced in the draft RIA, also indicated  an
interrelationship between sulfur control  and olefin content.
The magnitude of this interrelationship,  as developed in the
draft RIA, was somewhat greater than that developed from the B
& M data.   However, that data was only for one PADD  (PADD 3)
and therefore is less applicable to the national average than
the value developed here.  The fact that  the A/0 study  showed a
sulfur-olefin interrelationship for PADD  3 refineries while the
B & M model did not show an interrelationship in Region 3
demonstrates the variation between the assumptions contained in
different models.  Either relationship is possible, depending
on how the refineries decide to reduce sulfur levels.   It would
not be appropriate to "mix and match" data from  two different
models to develop a national average because of  such
differences.  Hence, since the cost data  from the B & M model
appeared to be reasonable, the results from B &  M for Region 3
were used.

     A national average factor was derived based on the impacts
shown in Regions 1 and 3.  Because of the uncertainties in the
costs for Region 2, the parameter data were not  included when
deriving a national average impact of reducing sulfur on

                               345

-------
olefins.   Factors representing the overall impact of reducing
the level of sulfur on the olefin content were developed from
the data in Table VI-10 for each of the three ranges of sulfur
reduction analyzed (down to 250 ppm, 250 ppm to 160 ppm, and
160 ppm to 50 ppm),  using the method outlined previously.  The
factors are listed in Table VI-7.

     Table VI-10 also shows that reducing the sulfur content of
gasoline typically leads to a reduction in aromatics.  This
impact is strongest in Region 2; for the reasons outlined
above, however, the data regarding the impacts of sulfur
control from Region 2 was excluded from the analysis.  In
Region 1, where olefins are reduced as a result of sulfur
removal,  one might expect the level of aromatics to increase to
provide additional octane; the opposite effect was observed.
In Region 3, where sulfur removal does not appear to impact
olefins,  one may expect there to be no change in aromatics,
since the volume of sulfur removed is too small to have much
impact on the concentration of aromatics (or any other
parameters)  remaining.  However, the impact indicated by the
data is small and is not unreasonable.  The data presented in
the draft RIA indicated no impact of reducing sulfur content on
aromatics.  However,  since the old data was based only on
estimates for PADD 3, and since the change in aromatics
resulting from a change in sulfur is seen in both regions in
the B & M data, it was determined that this interrelationship
should be included in the cost-effectiveness analysis.

     The overall impact of reducing the sulfur level on
aromatics over each of the three ranges analyzed was estimated
by weighting the percent change in aromatics and the real
change in sulfur in the manner described in subsection D.I.
The results of this calculation are presented in Table VI-7.

     The impact on E200 and E300 of a reduction in sulfur,  as
shown in Table VI-10, tends to be small and is somewhat
inconsistent between different regions.  Since there is no
easily identified change in refinery operations that would
likely result in a change in these parameters as a result of a
reduction in gasoline sulfur content, it was determined that no
interrelationship exists between sulfur and either of these
parameters.   This is supported by the Auto-Oil RFC study in
which reductions in sulfur had no consistent effect on E200 or
E300.

c.   Impact of Reducing Olefin Content on Other Fuel Parameters

     Olefin removal can be achieved through hydrogenation,
alkylation or oxygenate production.  Hydrogenation will also
reduce the level of sulfur, and valuable octane will also be
lost due to the reduction in olefins.  In the case of

                               346

-------
alkylation and oxygenate production, the loss of octane-rich
olefins would be offset by the addition of oxygenates or
alkylates, reducing the need to increase the severity of the
reformer to produce more aromatics.

     An analysis of the olefin interrelationships determined
from the B & M refinery modelling listed in Table VI-11
indicates a definite impact on E200, a minimal impact on
sulfur, aromatics, and E300, and no impact on other fuel
parameters.
     Data
                Table VI-11:
Used to Develop Olefins Interrelationship Factors
Controlled
Parameter
Olefins
Olefins
Olefins
Olefins
Reqion
1
2
3
1
2
3
1
2
3
1
2
3
From
7.2 vol%
5.3
8.9
6.9
8.9
6.9
7.2
5.3
8.9
6.9
8.9
6.9
7.2
5.3
8.9
6.9
8.9
6.9
7.2
5.3
8.9
6.9
8.9
6.9
To
6.9 vol%
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
6.9
5.0
Uncontrolled
Parameter
Sulfur
Aromatics
E200
E300
From
334 ppm
160
454
160
441
160
29.7 vol%
28.1
29.7
25.0
31.0
28.9
51.7 %evap
53 .3
57.4
53 .3
57.9
55.3
86.0 %evap
87.3
85.8
87.3
85.2
86.6
To
337 ppm
160
454
160
441
160
29.7 vol%
27.2
31.2
26.5
31.1
28.9
51.5 %evap
52.3
53.7
52.3
56.5
53.5
85.8 %evap
86.5
85.1
86.5
85.6
87.3
% Change
+ 2.0
0.0
0.0
0.0
0.0
0.0
0.0
-3.2
5.1
6.0
0.3
0.0
-0.4
-1.9
-6.5
-1.9
-2.4
-3.2
-0.2
-0.9
-0.8
-0.9
0.4
0.8
                               347

-------
     The data in Table VI-11 indicates no clear relationship
between a reduction in olefins and the sulfur level of the
fuel.  The only change in sulfur level was shown for Region 1
in reducing olefins down to 6.9%, and in this case sulfur
actually increased slightly.  Since this is not the expected
outcome of reducing olefins, and since the change in sulfur
level was very small, a relationship was not defined.  The data
used in the draft RIA indicated some impact on sulfur due to
olefins control, but the data was only for PADD 5 and the
magnitude of that impact was difficult to measure.  No
interrelationship between these parameters was assumed in this
cost-effectiveness analysis.

     The data in Table VI-11 also indicate that reducing the
olefins in the fuel has some impact on the aromatics content,
although the impact varies with the region and the level of
olefin reduction considered.  For example, in Region 1, where
there was a decrease in sulfur when olefins were decreased
(down to 6.9%), one would deduce that the olefins were removed
via hydrogenation and that aromatics content would, if
anything, increase to make up for the lost octane.  However
some isomerization of straight chain hydrocarbon gasoline
components to higher octane components that occurred may have
eliminated the potential aromatics effect.  In Regions 2 and 3,
where sulfur was not impacted by the reduction in olefins, one
would surmise that the olefins were reduced using alkylation
and that perhaps the aromatics content would be reduced, too.
But the data shows that the aromatics content of the fuels
produced in Regions 2 and 3 actually increase when the olefins
content is reduced.  Since each PADD was assigned a different
methodology for the reduction of olefins, it was not possible
to determine an interrelationship between olefins and aromatics
for this study.  Past studies indicated some change in
aromatics as a result of olefins control, but, again, this data
was based solely on PADD 5 refinery modelling.

     The data in Table VI-11 do show a relationship between
olefins and both E200 and E300, although the magnitude of these
relationships differs significantly.  Since the impact on E300
is  small and since the impact varied directionally in
different regions, this relationship was determined to be
insignificant.  However, the relationship between olefins and
E200 appeared to be more significant and was consistent across
the different regions.  In addition, it is reasonable to
surmise that E200 would decrease as the C4 and C5 olefins were
alkylated to heavier components.  Hence, this interrelationship
was included in the cost-effectiveness analysis.  The factor
representing the overall impact of reducing olefins on E200,
derived by weighting the results from each of the three
regions, is listed in Table VI-7.
                               348

-------
d.   Impact of Reducing RVP on Other Fuel Parameters

     The Reid vapor pressure (RVP)  of gasoline is reduced by
adjusting the level of butane and other hydrocarbons (primarily
pentane) in the fuel.  Most reductions in RVP will be achieved
by removing butane, however, deep cuts in RVP may require
removal of pentanes as well.  Both of these hydrocarbons are
high in octane and their removal would require that the lost
octane be replaced in the finished gasoline.  Octane may be
added by increasing  aromatics or oxygen, or by producing
additional alkylates (and using the displaced C4 and C5
hydrocarbons in the process).

     As shown in Table VI-12,  a reduction in RVP appears to have
some impact on aromatics, E200, and E300, but the size of the
impact and the agreement between different regions varies.
From the standpoint of refinery operations, there is no clear
connection between a reduction in RVP and the aromatics content
of the finished gasoline.  The impact shown here is small, and
can be considered insignificant given the small level of RVP
reduction considered.  Other studies show no impact on
aromatics from RVP control.  Similarly, the impact on E300 of
reducing RVP is small,  and the results are inconsistent between
the three regions, so no interrelationship was assumed.
However, the trend of E200 increasing with a reduction in RVP
is sizeable, and can be explained by the fact that removing C4
(and perhaps C5) paraffins would decrease E200.  Hence, an
interrelationship between RVP reduction and E200 was assumed in
the cost-effectiveness analysis.
                               349

-------
 Table  VI-12:  Data  Used  to  Develop  RVP  Interrelationship  Factors
Controlled
Parameter
RVP
RVP
RVP
Region
1
2
3
1
2
3
1
2
3
From
7.6 psi
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
7 .6
7 .3
6 .9
To
7.3 psi
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
7.3
6. 9
6.5
Uncontrolled
Parameter
Aromatics








E200








E300








From
29.7 vol%
29 .8
29 .7
28 .4
28 .5
28 .5
30 .5
30 .6
30 .7
51.1 %evap
50 .7
50 .5
55 .9
55 .2
55 .0
57 .4
56 .3
55 .6
85.8 %evap
85 .7
85 .6
86 .8
86 .7
86 .7
86 .6
86 .3
86 .5
To
29.8 vol%
29.7
29.7
28.5
28.5
27. 9
30.6
30.7
30.5
50.7 %evap
50.5
49.5
55.2
55. 0
53 .3
56.3
55.6
54 .5
85.7 %evap
85.6
85.5
86.7
86.7
86.4
86.3
86.5
87.7
% Change
0 .3
-0.3
0 .0
0 .4
0 .0
-2 .1
0 .3
0 .3
-0.8
-0.8
-0.3
-2 .1
-1.2
-0.4
-3 .1
-1. 9
-1.3
-2 . 0
-0.1
0 .0
-0.2
-0.1
0 .0
-0.3
-0.3
0 .3
1 .4
e.   Impact of Increasing Oxygen on Other Fuel Parameters

     Adding oxygen to gasoline increases the octane of the fuel;
other octane-rich components can then be reduced to produce
fuel with the desired level of octane.   (Some of this reduction
occurs simply via dilution of the other components.)  Since
many high-octane components of gasoline require severe
processing conditions to produce, a reduction in the amount of
                               350

-------
these compounds required is beneficial to the refinery.  It is
most likely that the addition of oxygen to the fuel would be
offset by reducing the aromatics content through a less severe
reforming configuration.  It is also possible that other fuel
parameters could be affected if the refiner elects to reduce
the volume of crude throughput to produce a constant volume of
products.  This would likely lead to a reduction in sulfur and
olefins in the finished gasoline due to changes in the
operation of the refinery (primarily the catalytic cracking and
reforming operations) in response to the lower volume of crude
processed.

     The Bonner and Moore refinery modelling data used to
estimate the impact of adding oxygen on other fuel parameters
is presented in Table VI-13, and indicates that in all three
regions both aromatics and E200 are affected by the addition of
oxygen.  The aromatics content of the fuel is reduced, which is
the expected response to the increase in octane provided by the
oxygenate.  However, this reduction is less than expected and,
as stated earlier, is probably caused by the large decrease in
olefins observed.  The change in fuel composition is expected
to result in a minor change in E200, as the data indicates.
However, the impact on E300 is small and is inconsistent
between regions.
                          Table  VI-13:
     Data Used to Develop Oxygen Interrelationship Factors71
Controlled
Parameter
Oxygen
Oxygen
Oxygen
Region
1
2
3
1
2
3
1
2
3
From
2.0 wt%
2 . 0
2 . 0
2 . 0
2 . 0
2 . 0
2 . 0
2 . 0
2 . 0
To
2 .7 wt%
2 .7
2 .7
2 .7
2 .7
2 .7
2 .7
2 .7
2 .7
Uncontrolled
Parameter
Aromatics


E200


E300


From
29.0 vol%
26 .3
30 .3
50.3 %evap
53 .0
54 .9
85.9 %evap
87 .2
86 .6
To
28.1 vol%
25 .0
28 .9
50.6 %evap
53 .3
55 .3
85.8 %evap
86 .5
87 .3
% Change
3 .1
4 .9
4 .6
0 .5
0 .7
0 .6
-0.2
-0.9
0 .8
 Based on data from Bonner & Moore for EPA.
     A review of the Auto-Oil data  (presented in Table VI-14
below),  which was developed by a different model, indicates
that an increase in oxygen results  in a large decrease in
aromatics.  This response of the refinery is more expected than
the one predicted by the B & M model.  However, the Auto-Oil
                               351

-------
data was developed only  for  a  single PADD,  and thus is not
directly applicable  to a national average estimate.

     Since it is reasonable  to expect that  many refiners will
want to have a significant reduction in aromatics upon addition
of oxygen, the results from  the two data sets were averaged,
and the average factors  used in this analysis.  Since both
studies also indicated some  interrelationship between oxygen
and E200, the average of this  impact from the two data sets
also was used in the cost-effectiveness analysis presented
here.  However, although the Auto-Oil data  shown in Table VI-14
indicates a relationship between oxygen and E300, such a
relationship was not included  in this analysis because the
Auto-Oil data was only based on PADD 5 and  the Bonner and Moore
data for the rest of the country did not strongly corroborate
such a relationship.
        Table VI-14:  Auto-Oil Data Used to Develop  Oxygen
        Interrelationship Factors - Based on PADD V Data71
Controlled
Parameter
Oxygen
From
0.4 wt%
To
2.7 wt%
Uncontrolled
Parameter
Aromatics
E200
E300
From
34.4 vol%
43.5 %evap
79.4 %evap
To
23.6 vol%
52.4 %evap
83.1 %evap

     from the Auto-Oil values for T50 and T90 using the conversion equations developed by EPA
     (see section III).
f.   Impact of Lowering Aromatics on Other Fuel Parameters

     Reducing aromatics significantly below the level achieved
through addition of oxygenates will require refining changes to
make up for the octane loss.  The octane could be increased
through increased alkylation of C4 and C5 olefins.  The reduction
in aromatics combined with a conversion of olefins could have an
impact on E200 and E300 depending on the changes in distillation
resulting from the new composition; these changes would be
refinery specific.

     The following table presents the data obtained from the B &
M refinery modelling to explain the relationship between
aromatics and other fuel parameters.  Although a change in
olefins would be expected, the results from the three regions
modelled are very inconsistent, and a trend cannot be determined.
The impact of aromatics reduction on E200 similarly is
inconsistent and makes it impossible to determine an
interrelationship between the parameters based on this data.
However, the results for E300, albeit small, are consistent and
make sense directionally, as it is reasonable to assume that E300
                               352

-------
may increase if the heavier aromatics are removed or reduced.
Hence,  an interrelationship between aromatics and E300 was
included in the cost-effectiveness analysis.
Data Used to Develop
                          Table VI -15:
                         Aromatics Interrelationship Factors
Controlled
Parameter
Aromatics
Aromatics
Aromatics
Reqion
1
2
3
1
2
3
1
2
3
From
30 vol%
28
24
28.3
28
24
30
28
24
30
28
24
28.3
28
24
30
28
24
30
28
24
28.3
28
24
30
28
24
To
28 vol%
24
20
28
24
20
28
24
20
28
24
20
28
24
20
28
24
20
28
24
20
28
24
20
28
24
20
Uncontrolled
Parameter
Olefins








E200








E300








From
5.70 vol%
6 .64
6.56
6.9
6 .9
6 .9
6.9
6.9
6 .9
51.3 %evap
53.2
55.3
55.8
55.4
54.7
62.8
58.7
59.1
85.9 %evap
86.7
88.2
86 .8
86.8
88.3
86 .3
87.6
88.5
To
6.64 vol%
6 .56
6.57
6.9
6 .9
6 .9
6.9
6.9
6 .9
53.2 %evap
55.3
57.7
55.4
54 .7
57.2
58.7
59.1
61.2
86.7 %evap
88.2
89.1
86 .8
88.3
89.4
87.6
88.5
90.2
% Chanqe
16 .5
-1.2
0.2
0.0
0.0
0.0
0.0
0.0
0.0
3 .7
3.9
4.3
-0.7
-1.3
4.6
-6.5
0.7
3 .6
0.9
1.7
1.0
0.0
1.7
1.2
1.5
1.0
1.9
g.   Impact of Increasing E200 or E300 on Other  Fuel  Parameters
                               353

-------
     Increasing the percent of fuel evaporated at either 200 or
300 degrees Fahrenheit requires a change in the fractionation
scheme at the refinery.  E300 control will require elimination
of many of the heaviest components of gasoline, while E200
control could result in removal or conversion of mid-range
components such as olefins.  These changes could have an impact
on the overall octane value of the fuel, and could thus have
implications for other fuel parameters.

     The data used to develop the interrelationship between E200
and other fuel parameters is listed in Table VI-16; this data
was generated in-house using the B & M refinery model.102  The
results from modelling Region 3 were somewhat questionable
since E200 started at a higher level than expected.  However,
because no other data was available to evaluate E200, the data
from all three Regions was used to develop the
interrelationships between E200 and other parameters.
     102Lester Wyborny,  November 4,  1993.

                               354

-------
    Table
Factors
VI-16: Data Used to Develop E200 Interrelationship
Controlled
Parameter
E200
E200
E200
Reqion
1
2
3
1
2
3
1
2
3
From
50
%evap
54
57
53
54
57
55
58
61
50
54
57
53
54
57
55
58
61
50
54
57
53
54
57
55
58
61
To
54
%evap
57
60
54
57
60
58
61
64
54
57
60
54
57
60
58
61
64
54
57
60
54
57
60
58
61
64
Uncontrolled
Parameter
Olefins
Aromatics
E300
From
5.14 vol%
5.45
5.14
6.41
7.23
8.26
8.44
8.90
8.90
29.2 %vol
26.9
23 .3
24 .9
25.3
23 .6
31.0
33 .6
32 .7
88.1 %evap
89.1
90.1
89.3
89.4
90.2
87.7
88.1
89.0
To
5.45 vol%
5.14
7.07
7.23
8.26
8.46
8.90
8.90
8.90
26.8 %vol
23.3
21.7
25.3
23.6
22.2
33.6
32.7
29.7
89.1
%evap
90.1
91.8
89.4
90.2
90.8
88.1
89.1
89.8
% Change
6.0
-5.7
37.5
12.8
14.2
2 .4
5.5
0.0
0.0
-8.0
-13 .2
-6.8
1.7
-6.7
-6.0
8.1
-2.7
-9.2
1.2
1.1
1.8
0.2
0.8
0.7
0.5
1.0
0.9
     The data in Table VI-16 shows some interrelationships
between E200 and the parameters aromatics and E300.  The data
for the relationship between E200 and olefins is inconsistent
across the three regions examined, and thus no
interrelationship between these parameters was determined.  In
contrast, the data shows a more significant relationship
                               355

-------
between E200 and aromatics,  although this is a somewhat
unexpected reaction  of  the  refineries to increasing E200.
Similarly, the data  shows a definite and not altogether
unexpected relationship between E200 and E300.

     The B & M refinery model  did not yield satisfactory results
for the control of E300.  Hence,  it was necessary to use the
data generated by Turner and Mason for Auto-Oil that was used
in the development of parameter relations for the draft RIA.
The Turner and Mason data looked at controlling T-90, the
temperature at which 90% of the fuel has evaporated.  Since the
Complex Model now includes  E300,  the T-90 data was transposed
to represent E300 based on  the relationship between these two
parameters cited previously.   This data is presented in Table
VI-17.  One weakness in this data is that it only represents
one region of refineries (PADD 3)  rather than modelling several
regions covering the entire country.  However, it does
represent 55 percent of the RFC likely to be produced and
provides one measure of how a  refinery may react to a need to
increase E300.
    Table VI-17: Data Used to Develop E300 Interrelationship
Factors
Controlled
Parameter
E300
E300
E300
Reqion1
PADD 3
PADD 3
PADD 3
From
84 %evap
88
91
84
88
91
84
88
91
To
88 %evap
91
96
88
91
96
88
91
96
Uncontrolled
Parameter
Olefins
Aromatics
E200
From
12.2 %vol
12 .6
12.4
20 %vol
20
19.5
52 . 9 %evap
55.6
56 .6
To
12.6 %vol
12 .4
11.5
20 %vol
19.5
17.3
55.6 %evap
56.7
58.6
% Change
3 .3
-1.6
-7.3
0.0
-2 .5
-11.3
5.1
1.8
3 .5
'This data is based on the Turner and Mason work for Auto-Oil.  For that study, refineries were
modelled in PADD-specific groups.  PADD 3 would correspond to the Bonner and Moore Region 3.


     The data presented above  indicates no clear relationship
between E300 and either olefins or aromatics,  and a measurable
relationship between  E300  and  E200.   The data is questionable
for the impact on  olefins, because for the first increase  in
E300 olefins increased,  while  for the second two increases,
olefins decreased.  The aromatics data is questionable because
in the Turner and  Mason work aromatics was limited to a level
of 20 vol%, and  appears to have reached this ceiling.  Hence,
                               356

-------
the data may not accurately reflect any interactions between
E300 and aromatics.  The E200 data shows the expected
correlation with E300, and therefore this interrelationship was
included in the cost-effectiveness analysis.

E.    Determination of Emissions Reductions

     The cost-effectiveness of a particular fuel component
change is determined based on reductions in in-use emissions
(i.e., the impact the fuel will have on emissions from vehicles
in-use at the time the fuel is sold).   In contrast,  the actual
Phase II reformulated gasoline emission performance standards
and NOx standards for RFC are based on the reformulated fuel's
effect on 1990 model year vehicle emissions relative to those
with Clean Air Act baseline fuel.  The following discussion
explains how baseline emissions and emission reductions from
baseline were determined.

1.    Non-Fuel Factors Affecting Baseline Emissions Determination

     The determination of both statutory baseline emissions and
in-use baseline emissions is affected by many factors, such as
specific characteristics of the fleet (age of vehicles, types
of vehicles, etc.) or the other regulatory programs in place at
the time (e.g., enhanced inspection and maintenance, fuel
volatility control).   The primary factors affecting the base
emissions used in this analysis are briefly discussed below.

a.    Enhanced Inspection and Maintenance

     Enhanced inspection and maintenance (I/M)  programs are
expected to be fully implemented and phased in by the year
2000.  The program is mandated for all serious and worse ozone
nonattainment areas.   Under enhanced I/M, emissions are lower
relative to a basic I/M program because vehicles are kept in
better condition.  Vehicles with poorly maintained exhaust and
evaporative emission control systems are identified and must be
repaired.  The impact of enhanced I/M programs on the emission
reductions achieved is that the same percentage reduction in
emissions applied to a lower baseline emissions value yields
lower absolute reductions.  Enhanced I/M is assumed in this
analysis.

b.    MOBILE Model

     The MOBILEBa emissions model (with enhanced I/M)  is
utilized in this analysis to determine the in-use baseline
emissions.   Different base emissions would result if earlier
versions of the MOBILE model were used.   Baseline emissions
determination is also affected by the average vehicle speed
used in the MOBILE model.  In this analysis, EPA utilized its

                               357

-------
standard Federal Test Procedure (FTP)  speed of 19.6 miles per
hour.  EPA is currently investigating other speeds or
distributions of speeds to more accurately estimate baseline
emissions.  Revised speed estimates would likely be higher,
decreasing VOC and increasing NOx emissions estimates.
However, since the investigation is not yet complete, the FTP
value was assumed here.

     One comment received on the proposal raised concern with
using the MOBILES model instead of the MOBILE4.1 model.  The
exhaust emissions calculated by MOBILES are significantly
higher than those calculated by MOBILE4.1.  The commenter
suggested that the cost-effectiveness analysis presented by EPA
was optimistic because it was based on a percent reduction in
emissions applied to this larger emission inventory, in effect
improving the cost-effectiveness of each incremental change in
fuel composition because the number of tons over which the
costs were spread was large.  MOBILES (and the subsequent
MOBIlESa model)  went through a thorough review process,
including a number of public workshops.   In particular, EPA
believes that the increase in in-use VOC exhaust emissions
relative to MOBILE4.1 are real.  Non-exhaust VOC emissions did
not increase much in MOBILES relative to MOBILE4.1, because
time and resources did not allow a number of possible
modifications in the nonexhaust emission methodology to be
implemented.  Future modifications to the MOBILE model could
produce even higher nonexhaust VOC emissions.  In any event,
the VOC emissions reductions on which the analysis is based are
reasonable.  The MOBILES model contains the latest information
on vehicle emissions and fuel economy.  It is the most
realistic model available to estimate the fleet emissions in
the future that is available.

c.   Other Factors Influencing Emissions

     Other factors can also influence base emissions and thus
the reductions determined from those base emissions.  For
instance, Stage II evaporative controls (controls at the fuel
pump to reduce evaporative emissions when refueling vehicles)
are included in the MOBILESa analysis.  Onboard (on the
vehicle) refueling emissions controls are not considered,
however, since the rules governing them have not been
promulgated and such controls were not installed on 1990 model
year vehicles.   The incremental control of onboard vapor
recovery to Stage II is small compared to the total vehicle VOC
emissions where both are in place.  Given the current court
order under which EPA is working  (see 58 FR 30731), it is
likely that vehicles sold in the year 2000 and/or later will
have onboard refueling control systems.   For the purposes of
this analysis,  however, this would not greatly affect the
                               358

-------
baseline in-use emissions because fewer than 50% of the
vehicles would likely have such controls at that time.

2.    Baseline Fuel Assumptions

     Changes in fuel composition and emissions relative to the
Clean Air Act baseline fuel are used to determine the VOC, NOx,
and toxics emissions performance standards.  The base fuel
composition from which the cumulative emissions reductions were
estimated (and thus from which the proposed performance
standards were determined) was the statutory summertime
baseline gasoline presented in section 211(k)(10) of the Act.
The baseline emissions are estimated using the Phase II complex
model as discussed in section III.

     In the draft RIA, northern area (VOC control region 2)
emission reductions were determined relative to the statutory
baseline.103  Southern area (VOC control region 1) emission
reductions were shown relative to both the statutory baseline
and to a baseline with all of the statutory values except for
RVP, which was set at 7.8 psi (the 1992 Phase II gasoline
volatility requirement for southern ozone nonattainment areas).
The proposed standards for southern areas were determined
relative to the latter baseline.  Several comments were
received expressing a concern about setting standards for the
two regions relative to different baselines.  In this revised
analysis, the standards for both VOC control regions are
determined relative to the statutory baseline RVP of 8.7 psi.

     Based on comments received and on its own analysis, EPA
realized that the actual 1990 summertime olefin value was
approximately 13 volume percent (vol%)  while the value
specified in the Act was 9.2 vol%.  This difference between the
actual 1990 value and the statutory value significantly impacts
estimations of emissions, total reductions in emissions, costs,
and cost-effectiveness.  Additionally,  the actual impact on
refiners of utilizing an unrepresentative baseline in setting a
performance standard could result in anti-competitive impacts.
Therefore, in determining cost-effectiveness,  EPA used a
nationwide summertime baseline olefins value of 13.1 volume
percent.  This value was determined from the Bonner & Moore
modelling work, and includes 1990 olefin levels in California.
The value is in agreement with EPA's own estimate based on
gasoline surveys in select cities across the country.  The
baseline emissions when the olefin content is 13.1 volume
percent  (vol%) and when the olefin content is 9.2 vol% are
shown in Table VI-18 below for comparison.  Once the percent
     103In the draft RIA,  southern areas were called "Class B"
and northern areas were called "Class C."

                               359

-------
reduction standards  are  determined based on the cost-
effectiveness of  the fuel  changes relative to the in-use
emissions assuming the  13.1 vol% olefin level, refiners must
still certify their  fuel relative to the statutory baseline
fuel containing 9.2  vol% olefins.104
                          Table VI-18:
   Phase II RFG Emissions  at  Different Olefin Levels  (mg/mi)71
VOC Control
Region
Olefin Level
VOC
NOx
Benzene
Acetaldehyde
Formaldehyde
1 , 3 -Butadiene
POM
Region 1 (Southern)
9.2
1466
1340
52
4
8
7
3
13.1
1456
1369
52
4
7
8
3
Region 2 (Northern)
9.2
1399
1340
52
4
8
7
3
13.1
1389
1369
52
4
7
8
3
^Determined using Phase II complex model with all other fuel parameters at
Clean Air Act summer baseline values.
     To determine the cost-effectiveness of Phase  II  RFG
incremental to Phase I, the costs and emissions  reductions
associated with the Phase I fuel had to be evaluated.  Hence,
while the Phase II standards are expressed relative to the  Clean
Air Act baseline, the cost-effectiveness of fuel reformulations
in excess of the Phase I requirements were determined relative to
a fuel meeting the Phase I definition  (2 wt% oxygen,  1 vol%
benzene, and 8.0 psi RVP).

     Because the Phase II VOC and the NOx emission reduction
requirements are only required during the high ozone  season,
there are no Phase II or NOx control requirements  outside of  the
high ozone season.  Phase II toxics control applies year-round.
However, toxics control has been shown to be cost-ineffective
(see subsection F.3 below as well as the draft RIA),  and thus was
     104The Phase I complex model  indicates that refiners will be
able to meet the Phase  I  requirements (15% reduction  in VOC,  no
NOx increase) even with an  olefins level of 13.1%.
                               360

-------
evaluated using the same costs and emissions baselines that were
developed for VOC and NOx control.  Hence,  there is no need for a
"wintertime" baseline fuel for Phase II RFC (although the
wintertime baseline fuel still applies for the antidumping
provisions; see section III).

3.    In-Use Basis

     The performance standards developed in this document are set
as percentage reductions in emissions from the statutory baseline
fuel, as described above.  However, in-use emissions better
estimate the emissions of vehicles that will be on the road
during and after the year 2000.  Use of the statutory baseline to
estimate cost-effectiveness would only represent the emissions
reductions that would be achieved from 1990 technology vehicles.
Thus it is more appropriate to base the cost-effectiveness, i.e.,
the cost per actual benefit,  on the emissions reduction from the
in-use emissions baseline.

     Baseline exhaust and nonexhaust emissions of vehicles in-use
in the year 2003 (a representative year after 1999) were obtained
from MOBILE5a,  assuming enhanced inspection and maintenance
programs in all severe or worse ozone nonattainment areas.  This
emissions estimate includes light-duty gasoline vehicles and
trucks.  To determine the in-use emissions resulting from each
fuel parameter change, the in-use baseline emissions were
adjusted by the percent change in the Phase II complex model
emissions projections calculated from that fuel parameter change.

4.    Emission Reductions

     Incremental reductions in emissions are determined relative
to emissions from a previous fuel formulation, while total
reductions are determined relative to the emissions resulting
from Phase I RFC and also to the emissions from the statutory
baseline fuel.   This allows calculation of the cost-effectiveness
based on the incremental reduction in in-use emissions.  It also
permits determination of an emission reduction standard based on
the total reduction from the statutory baseline emissions.

     To determine each incremental reduction in emissions, the
emissions reduction (calculated in grams/mile) must be expressed
in terms of the net tons emitted.  (The grams/mile numbers can be
used directly to calculate reductions in the number of cancer
incidences as a result of toxics control, as discussed in section
F below).  Estimates of the level of fuel consumption and the
average fuel economy of the fleet are required to complete this
calculation.

     Summer fuel consumption estimates in the year 2000 for ozone
nonattainment areas in VOC control regions 1 and 2 were estimated
based on the 1990 fuel consumption and population data from the

                               361

-------
1990 census for the mandated RFC areas, the currently opted-in
areas,  and for the remainder of the Ozone Transport Region.  The
National Petroleum News (June 1992) showed 1990 fuel consumption
to be approximately 115 billion gallons.105  A DOE fuel
consumption projection for the year 2000 is also about 115
billion gallons.106  Since  there was little difference between the
actual 1990 and projected 2000 fuel consumptions, EPA used 115
billion gallons per year as its base number.  The ratio of the
population of an RFC area to its state population was then
multiplied by the state summer fuel consumption  (from May-
September 15--again, estimated from 1990 actual fuel consumption
data).   Summing these estimates for individual states yielded the
total summer fuel consumption for all the RFC areas.

     A fleet fuel economy of 20.58 miles per gallon was estimated
from the MOBILE4.1 fuel consumption model for the year 2000 and
was used in this conversion calculation.  This fuel economy
includes both light-duty vehicles and light-duty trucks.  The
MOBILE4.1 model was used because a fuel consumption model based
on MOBILESa has not been developed.

     Finally, in order to provide an equitable comparison of the
proposed reformulated gasoline VOC and NOx controls with other
VOC and NOx control programs, the incremental VOC and NOx
emission reductions, in tons, were annualized, i.e., converted
from summer tons to annual tons.  One commenter recommended that
cost-effectiveness be determined on the basis of actual summer
tons.  While it is true that the RFC program is a summer program,
and although ozone formation primarily occurs during summer
months,  historically EPA has measured the cost-effectiveness of
VOC control programs in terms of annual tons reduced.  This
precedent has been set by stationary source controls, which
operate year-round but only achieve ozone benefits in the summer
months,  as well as mobile source controls, such as the recent
gasoline volatility regulations.  In annualizing the emission
reduction benefits, EPA assumed a summer season of 5.5 months (to
account for the lead time needed to produce and deliver RFG).
This summer season is longer than the one used in the draft RIA
(4.5 months).  It was adjusted in response to a comment received
which stated that the total time over which refiners must produce
reformulated gasoline was understated in the draft RIA  (as
discussed previously).

     A similar analysis methodology was followed for estimating
toxics emissions reductions.  Baseline values of each toxic were
     105reference  (chris b.)?

     106Energy Information Administration,  Annual  Energy Outlook:
1993. DOE/EIA-0383(93).

                               362

-------
obtained from the Phase II complex model.  New emissions were
determined from the complex model for each fuel component change.
The emission reductions were determined from the sum of the
toxics emissions and was determined relative to the previous fuel
composition.  Thus, tons of toxics reduced are determined from
the sum of all of the toxics emissions.
F.   Cost-Effectiveness Determination

     In this analysis, cost-effectiveness is defined as the cost
of an emission control strategy per ton of VOC or NOx emissions
reduced or per cancer incidence (CI)  avoided (for air toxics).
Cost-effectiveness can be used to compare different control
strategies or different options within one control strategy
(e.g.,  incremental changes in fuel composition).  A low cost per
ton reduced (or per cancer incidence averted)  is desirable.

     In general,  for this analysis EPA estimated the incremental
emissions reductions associated with a given fuel component
change, and determined the incremental cost-effectiveness for
that fuel change relative to the previous fuel composition.  A
copy of the results of this analysis is attached as Tables VI-A5
and VI-A6 in the Appendix.  The tables report incremental cost-
effectiveness for reductions in VOC,  NOx, VOC + NOx (and several
related VOC + NOx cost-effectiveness expressions) and toxics for
each of the two VOC control regions.

1.   Comments on Analysis in Draft Regulatory Impact Analysis

     In the draft RIA, EPA presented an analysis of the cost-
effectiveness of VOC, NOx, and toxics control based on the
refinery costs and emissions reductions estimated in that study.
The costs were evaluated in terms of the reductions achieved,  and
were also expressed on a per-gallon-gasoline-produced basis to
more directly reflect the costs to consumers.   EPA received
comments on many parts of the analysis.  One commenter suggested
that EPA analyze the cost which would be incurred by each
household to bring an area into compliance with the NAAQS, and
use this as the basis for the cost-effectiveness analysis.  While
such an analysis may be interesting from an economic standpoint,
it was not required for EPA to perform this type of analysis to
determine the cost-effectiveness of the reformulated gasoline
program.  The methodology used in the draft RIA and here is
consistent with the methodology used to analyze and support all
other stationary and mobile source emission control programs.

     Several commenters presented evaluations of the cost of
Phase II RFG over Phase I, based on the proposed standards.
These evaluations suggested costs which ranged from 5.5-15C/gal,
depending on the level of VOC control  (25-35% reduction).  It was
estimated by another commenter that the cost for small refiners

                               363

-------
may be 5C/gal greater than for large refiners.  As will be
explained below, EPA has estimated that the cost of VOC control
will be significantly less than the range projected by the
commenters, depending on the level of control required and the
parameters controlled to achieve that standard.  However,  EPA
agrees that some refiners will realize costs higher than the
average estimates presented here.  A more detailed discussion of
the impacts of this rulemaking on small entities is presented in
section VIII of this document.

     Another comment emphasized the need to use a "valid"  complex
model when determining the cost-effectiveness for use in setting
the Phase II standards.  As discussed above, EPA used the final
complex model to determine the emissions reductions achievable
through incremental changes in fuel parameters.  These emissions
reductions will be the basis of the cost-effectiveness analysis
presented below.

2.   Factors Affecting Cost-Effectiveness Determination

a.   Phase I Fuel Requirements

     In this analysis, the combined effect of the Phase I RFC
requirements was evaluated prior to evaluating the effects of
additional fuel parameter changes because the costs of these
requirements will be incurred by refiners prior to the year 2000.
After accounting for the production of Phase I RFC,  the most
incrementally cost effective fuel changes were made.  The
analysis proceeded through each potential fuel change until the
most costly (highest cost per reduction achieved) parameter
change was identified.

b.   Complex Model Fuel Parameter Range Limits

     The range over which fuel components were allowed to change
was governed by the data from which the fuel component control
costs were derived, with the additional check that the limits
defined by the complex model for each fuel component were not
exceeded.  If cost data was available for several levels of
control of a single fuel component, each level of control was
incorporated as a discrete control level, as described
previously.  The fuel components and the specific levels of
control to which each fuel component was held are discussed in
section IV.

c.   VOC Control Region

     Separate cost-effectiveness analyses must be done for each
VOC control region because of differences in the nonexhaust
emissions experienced in the two regions (resulting primarily
from the different average ambient temperatures) which affect the
total percent and gram per mile reductions for VOC,  NOx, and

                               364

-------
toxics.  Total percent reductions are determined from the CAA
statutory summer baseline gasoline composition.

3.   Analysis Methodology

     For the purpose of this analysis,  cost-effectiveness is
defined as the ratio of the incremental cost of a control measure
to the incremental benefit, e.g., the cost in dollars per ton of
VOC or NOx emissions reduced or the number of cancer incidences
avoided.  The use of cost-effectiveness allows for the relative
ranking of various control strategies so that a specified
environmental goal can be achieved at minimum cost.

     In this analysis, cost-effectiveness was evaluated using
incremental fuel parameter controls and costs.  In other words,
only a single fuel parameter was modified at a time, e.g., RVP
was reduced to 7.3 psi from 7.8 psi,  or sulfur was reduced from
250 ppm to 160 ppm.  The incremental fuel component control cost
(in cents per gallon)  determined as described in section C above
was converted to total incremental cost (in millions of dollars)
using the same average fuel consumption data that was used to
calculate total emissions.  The incremental cost-effectiveness
was determined by dividing the total incremental cost by the
actual incremental reduction in emissions (in tons or in cancer
incidences).   Cumulative actual reductions,  total cost, and total
cost-effectiveness are determined in an analogous manner.
Cumulative numbers are identified as "total."

     Determining the cost-effectiveness of toxics reduction
requires a slightly different approach.  Reductions in cancer
incidence must be determined separately for each toxic, and then
the resulting cancer incidences added together to determine the
overall reduction in cancer incidence for each year that the
program is in place.  Thus, for each toxic air pollutant,
emissions reductions  (in mg/mile) were converted to cancer
incidence per year  (Cl/yr) using a "unit risk factor" for each
toxic and the population of the participating nonattainment area.
Each unit risk factor (cancer incidences-per-year per gram-per-
mile per person) was assumed to be constant.  In the draft RIA,
unit risk factors were derived from an EPA paper of 1989.  For
this analysis, updated unit risk factors were derived from data
presented in EPA's recent Air Toxics Study107.   The factor for
converting gram per mile emissions to micrograms per cubic meter
used in the Air Toxics Study was adjusted to reflect the carbon
monoxide emissions predicted by MOBILE5a instead of using the
factor derived from MOBILE4.1 described on page 4-7 of that
     107U.S. EPA, "Motor Vehicle-Related Air  Toxics  Study,"
Office of Mobile Sources, Report  No.  EPA  420-R-93-005,  April
1993.

                               365

-------
report.  The emissions were adjusted from 1988 to the year 2000
with the factor 1.218 to reflect anticipated changes in vehicle
miles travelled.  The adjustments for ambient conditions and
potency of each toxic were used as described in the Air Toxics
Study.  Table VI-19 presents the unit risk factors derived for
this analysis.
          Table VI-19: Unit Risk Factors for Air Toxics
Toxic Pollutant
Benzene
Formaldehyde
Acetaldehyde
1, 3 -Butadiene
POM
(CI/yr)/(g/mi)/
(person)
3.696 x 10"6
4.910 x 10"6
2.048 x 10"6
9.103 x 10"5
5.344 x 10"5
     The Phase II reformulated gasoline standards can be
determined from this cost-effectiveness analysis, by determining
the maximum level of control that can be achieved for a
reasonable incremental cost-effectiveness.  Other factors,  such
as the economic impact on the industry and the technological
feasibility of that level of control, can also be considered when
choosing the standard.

     EPA received many comments on the methodology used in the
draft RIA (which is essentially the same methodology used in this
analysis),  as well as suggestions for alternative methods of
determining cost-effectiveness.  One recommended method involved
using a refinery model consisting solely of Petroleum Allocation
for Defense District  (PADD)  1 refineries to develop a
relationship between cost and a single pollutant (e.g., VOC),  and
using this relationship to select an average standard based on a
cost-effectiveness limit.  According to this commenter, the cost-
effectiveness limit should be based on the "knee" in the cost
curve (the point at which costs escalate dramatically for an
incremental reduction in emissions), i.e., the maximum acceptable
cost-effectiveness should be determined individually rather than
being based on a preselected value.  The standards for other
pollutants should then be set based on the fuel composition used
to reach the VOC standard.

     The methodology used by EPA does not differ significantly
from the recommended one.  As explained above, the standards were
developed using the complex model to determine emissions
reductions.   Rather than basing the costs on a single region,  EPA
                               366

-------
averaged the costs for several regions, excluding California,
based on the relative volumes of RFC produced in each region.  In
addition to use of the cost-effectiveness limit, EPA also
attempted to insure that the standards provided adequate
flexibility for individual refiners, prevented typical refiners
from incurring unreasonable costs, and did not force the industry
to make a fuel with a specified formula.

4.   Results of Cost-Effectiveness Analysis

     Tables VI-20 and VI-21 present the results of the cost-
effectiveness analysis performed by EPA.  The per-gallon costs
and (average) cost-effectiveness for VOC and NOx reductions are
listed, based on the cost estimates for VOC control region 2
(northern areas).  The cost-effectiveness of VOC reductions for
VOC control region 1 will be slightly less than those shown in
Table VI-20.  The incremental cost for each parameter change will
be slightly less in region 1 because less RFG will be consumed
there,  and thus the overall cost will be less.  In addition,
fewer tons of VOC will reduced in the region due to differences
in the emissions predicted by the complex model for the two
regions, but the reduced cost overrides the $/ton analysis.  In
contrast, the cost-effectiveness of VOC control for control
region 1 compared to Phase I RFG will be greater than those for
control region 2.  This is largely due to the fact that the
actual emissions reductions achieved in region 1 will be lower
than in region 2 since these areas start with a lower RVP fuel
(as a result of EPA's fuel RVP regulations).   The incremental
cost-effectiveness of NOx control is very close between the two
regions, largely because the incremental reductions in emissions
(mg/mi) estimated by complex model are close.  Similarly, the
cost-effectiveness for NOx control incremental to Phase-I RFG is
quite close between the regions.  This is because the baselines
are closer than they were for VOC (since RVP has a negligible
impact on NOx).
                               367

-------
  Table VI-20:  Fuel  Parameter Control  Costs  and VOC  Reductions
                                                                       /i
Fuel Parameter
Control
Phase I
RVP: 8.0 psi
Oxygen: 2.1wt%
Benzene: 0.95%
RVP to 7.1 psi
RVP to 6.7 psi
Sulfur to 250 ppm
Sulfur to 160 ppm
Sulfur to 138 ppm
Sulfur to 100 ppm
Olefins to 8 . 0 vol%
Aromatics to 20vol%
Oxygen to 2 . 7 vol%
Olefins to 5 . 0 vol%
E300 to 88%
E300 to 91%
E200 to 44%
E200 to 47%
E200 to 50%
Incremental
Cost
(
-------
  Table VI-21: Fuel Parameter Control  Costs  and NOx Reductions
                                                               /i
Fuel Parameter
Control
Phase I
RVP: 8.0 psi
Oxygen: 2.1wt%
Benzene: 0.95%
RVP to 6.7 psi
Sulfur to 250 ppm
Sulfur to 160 ppm
Sulfur to 138 ppm
Sulfur to 100 ppm
Olefins to 8 . 0 vol%
Aromatics to 20vol%
Oxygen to 2 . 7 vol%
Olefins to 5 . 0 vol%
E300 to 88%
E300 to 91%
E200 to 44%
E200 to 47%
E200 to 50%
Incremental
Cost
(C/qal)

_
0.12
0.56
0.24
0.52
0.78
2.01
0.61
2 .77
0.35
2.01
0.38
1.32
2.97
Cumulative
Reduction
(%)

0.4
2.4
5.8
6.8
8.7
10.8
11.9
12.5
14.1
14.1
14.2
13.9
13.7
13 .5
Incremental
Cost-Eff .
($/ton)

_
1,300
3,700
5,200
6,200
8,000
40,000
25,000
37,000
(-)
820,000
(-)
(-)
(-)
Incremental
to Phase I
($/ton)

_
3,200
3,500
3,700
4,200
5,000
8,200
8,900
12,000
13,000
16,000
17,000
19,000
24,000
71 Based on costs and emissions reductions for VOC control region 2 (northern
areas).  Assumes all costs allocated to NOx control.
     As Table VI-20 shows,  the  largest  reduction in VOC emissions
can be achieved through reductions  in the RVP of the fuel.   These
reductions can be achieved  at a cost-effectiveness well below the
$5,000-$10,000/ton levels proposed  in the February NPRM.  Control
of sulfur and other parameters  can  result in incremental
reductions in VOC, but the  costs increase dramatically, thereby
reducing the cost-effectiveness.  Similarly,  Table VI-21 shows
that the largest reduction  in NOx emissions occurs as a result of
reducing the level of sulfur in the fuel.  Attempts to reduce NOx
emissions by controlling other  parameters are in most cases cost
ineffective and in some cases detrimental.

     Overall, controlling RVP to 6.7 psi  and sulfur to 250  ppm
could yield an overall reduction from baseline VOC emissions of
about 26% at an incremental cost-effectiveness of about $3,700
per ton of VOC reduced.  However, the total cost-effectiveness of
                                369

-------
this level of VOC reduction relative to Phase I RFC is only about
$600/ton.  This latter value is important when comparing the
cost-effectiveness of potential VOC standards with alternative
methods of VOC control (see section H),  since these other
programs are usually evaluated based on the cost-effectiveness
relative to the baseline rather than the cost-effectiveness of
the last increment of control.

     Reducing the sulfur level of the fuel further, to 138 ppm,
would yield an overall reduction in NOx emissions of 6.8% (plus a
total VOC reduction of 27.4%) at an incremental cost-
effectiveness of about $5,200 per ton of NOx (assuming that all
of the costs are allocated to the NOx control) ,108  Again,  if one
notices the cost-effectiveness incremental to Phase I RFG, this
level of NOx control is even more cost-effective at about
$3,700/ton.  The overall cost-effectiveness of this level of VOC
and NOx control relative to Phase I RFG, if each unit of NOx is
valued equally to each unit of VOC, is $4,100/combined ton  (see
Table VI-A6 in the Appendix).  If NOx reductions are deemed to be
more valuable than reductions in VOC (as some researchers have
suggested), the cost-effectiveness of this strategy improves.

     NOx reductions exceeding 10% could be achieved with more
stringent sulfur control (down to 100 ppm)  as well as control of
olefins, at incremental cost-effectiveness less than $10,000 per
ton of NOx.  However, it should be noted that the olefins control
is more costly than readily apparent due to an increase in VOC
emissions  (as predicted by the complex model).   Under such a
scenario, total VOC reductions would be 27.8%.

     Table VI-22 presents the results of EPA's analysis of the
cost-effectiveness of toxics reductions.  This information is for
reductions in total toxics  (the five toxics of concern for
reformulated gasoline).  The cost per cancer incidence is a
function of the reduction in each individual toxic, since each
toxic has a different risk factor for cancer, as discussed
previously.  As the table shows, toxics control in Phase II RFG
is not cost-effective when expressed either on a $thousand/ton
basis or a $million/cancer incidence basis.  Toxics control of
25% reduction would cost over $220,000 per ton of overall toxics
reduced for the last increment of reduction  (from 24.6% to
25.0%), or close to $40 million per cancer incidence avoided.
Less than one cancer incidence would be avoided nationwide per
year at a toxics reduction of about 25%.  The cost-effectiveness
     108If the cost-effectiveness of this  level  of  NOx reduction
is credited based on the additional level of VOC control
achieved, at a value of $500/ton VOC  reduced,  the  cost-
effectiveness of this last  increment  of NOx reduction is about
$5,000/ton NOx.

                               370

-------
of toxics control  in VOC control region 1 would be  even higher
(see Table VI-A5 in  the Appendix).
Table VI-22: Fuel  Parameter Control Costs and Toxics  Reductions
                                                                 /i
Fuel Parameter
Control
Phase I
RVP: 8.0 psi
Oxygen: 2.1wt%
Benzene: 0.95%
RVP to 7.1 psi
RVP to 6.7 psi
Sulfur to 250 ppm
Sulfur to 160 ppm
Sulfur to 138 ppm
Sulfur to 100 ppm
Olefins to 8 . 0 vol%
Aromatics to 20vol%
Oxygen to 2 . 7 vol%
Olefins to 5 . 0 vol%
E300 to 88%
E300 to 91%
E200 to 44%
E200 to 47%
E200 to 50%
Incremental
Cost
(
-------
effectiveness of controlling any one of the toxics, if all of the
costs were allocated to that control, range from several hundred
thousand to several million dollars per ton of pollutant reduced.


G.   Comparison with Other RFG Cost-Effectiveness Analyses

     Contrary to VOC and toxics cost-effectiveness, which are
either very economical or too costly, respectively, NOx cost-
effectiveness will figure substantially into the decision-making
process for establishing a NOx standard.  For this reason, NOx
cost-effectiveness values were compared between various refinery
models to better understand the cost of reducing NOx.  The sulfur
reduction cases (sulfur has the largest impact on NOx reductions)
from the B & M cost data were compared to the cost data from
other models which optimized around NOx reduction  (i.e., refinery
models into which the complex model was incorporated directly).
Even though multiple parameters changed in these other NOx
studies, sulfur was still the major factor which the refinery
models chose to modify.  This observation at least verifies that
the models are similar.

     An important reason for making the NOx comparison in
addition to the sulfur comparisons made above in section C is to
allow the comparison with additional studies which could not be
used specifically for the sulfur analysis.  The cost-
effectiveness analysis by DOE,109 and the Turner/Mason cost-
effectiveness analyses for the National Petroleum Council and API
could then be compared.  Several discussions were held with DOE
and with API late in 1993 concerning some of the assumptions made
by API in the running of their computer model.110  These
assumptions are important since they determine, to a large
degree, the difference in sulfur costs noted above.  The DOE
staff integrated some of these assumptions into their refinery
model.  In addition, these later runs allowed the DOE model to
incorporate the final version of the complex model (although
negligible differences would be expected between the final
version and other late versions on which the API model were
based).  To compare the NPC data, the parameter values for
several of the computer runs (case code numbers Q40L and Q6N
Base) were entered into the complex model and the resulting NOx
changes were noted.  The cost-effectiveness calculated from the
various refinery models were normalized to the EPA data on the
basis of the fuel economy of the vehicle fleet in 2003, the 2003
     109Wyborny, Lester, U.S. EPA,  "Documents  Concerning Refinery
Modeling for the Reformulated  Gasoline  Program,"  Memorandum to
the Docket.

     110Ibid.

                               372

-------
vehicle fleet NOx emission factors,  and the same methodology for
calculating cost-effectiveness (annualized tons, 10% rate of
return on investment, etc.).

     NOx cost-effectiveness was compared separately for PADD 1
and PADD 3 refinery model runs.  The plot of PADD 1 cost-
effectiveness is shown in Figure VI-A2 located in the appendix.
The DOE cost-effectiveness curve is steeper initially than the B
& M curve, showing a lower cost-effectiveness value (more cost-
effective) at the same percent NOx reduction.  The DOE curve
changes to a higher cost-effectiveness value after it crosses the
B & M line.  The DOE data corroborates the B & M data at one
point just above the 7 percent NOx reduction line, indicating
that greater levels of NOx reduction become less cost effective.
The API data parallels the B & M plot but indicates higher cost-
effectiveness values (i.e.,  less cost effective control) at each
percent reduction in NOx value.

     The plot of cost-effectiveness versus NOx reduction for PADD
3 is shown in Figure VI-A3.   This plot shows that the DOE,  Auto-
Oil and NPC data all corroborate the B & M data by showing
similar levels of cost-effectiveness in the range of 6-9%
reduction in NOx.  In contrast, the API data is significantly
higher in cost,  as it was for PADD 1.

H.   Factors to be Considered in Setting Phase II RFG Standards

     Various economic,  technological, and environmental factors
should be considered in addition to the estimated cost-
effectiveness when setting performance standards for Phase II
reformulated gasoline.   The following sections present EPA's
consideration of many of these issues.

1.   Comparison to Cost-Effectiveness of Other Control Strategies

     The cost-effectiveness of VOC and NOx control under the
Phase II reformulated gasoline program is most easily evaluated
when compared to other stationary and mobile source VOC and NOx
control strategies.  Several comments received on the proposed
standards recommended that EPA evaluate the cost-effectiveness of
the Phase II standards compared to other control measures.   One
commenter suggested that such a comparison should include the
option to adopt programs such as RVP control instead of
reformulated gasoline.   The draft RIA did contain some
comparisons with other mobile and stationary source control
programs.  That comparison has been expanded here to include a
more detailed discussion of the cost-effectiveness of alternative
means to control VOC and NOx emissions from stationary and mobile
sources.

     It is important to note that not all of the other studies
compared here used the same methodology used by EPA to determine

                               373

-------
the cost-effectiveness of Phase II RFC.  This analysis is based
on the incremental cost-effectiveness of a particular change in
fuel composition relative to a previous, less costly change.
Most other analyses determine overall cost-effectiveness relative
to a baseline value.  If multiple options are considered by these
studies,  the cost-effectiveness of each option is evaluated
relative to this baseline rather than relative to the preceding
option.  Hence, in many cases it is more appropriate to compare
the incremental cost-effectiveness of Phase II RFC relative to
Phase I RFC with the cost-effectiveness of other studies.

     Another factor which will affect the comparison of the cost-
effectiveness of Phase II RFC with other control programs is the
fact that the Phase II standards are being set now (as agreed in
Reg-Neg)  even though they will not take effect until the year
2000.  The cost-effectiveness of VOC and NOx programs being
evaluated seven years from now are expected to be much higher
than those estimated here, since many of the less costly controls
will already be implemented.  Although EPA cannot project what
controls may be under consideration by the year 2000, it is
reasonable to assume that the Phase II RFC program will be
competitive by comparison.

a.   Stationary Source Emission Control Cost-Effectiveness

     In response to the cost-effectiveness analysis presented in
the proposal, one commenter stated that since EPA used average
costs which may be low compared to the costs realized in PADD 1
(since PADD 1 will have a high fraction of RFG produced), the
cost-effectiveness of Phase II RFG in PADD 1 would greatly exceed
the cost-effectiveness of stationary source controls in PADD 1.
EPA disagrees,  based on the cost-effectiveness estimates
presented below.  Furthermore, comments received from the state
regulators in PADD 1 indicated that the cost-effectiveness of
Phase II RFG presented in the draft RIA was competitive with the
cost of various control strategies currently in use or planned.

     The cost-effectiveness of stationary source VOC emissions
was determined from recommended control strategies for severe
nonattainment areas.111  A review of the estimated cost-
effectiveness of controlling VOC emissions from stationary
sources yielded a wide range of values.112  Many existing VOC
     11:LThe Clean Air Act section 183 (d) Guidance  on Cost-
Effectiveness; OAQPS, U.S. Environmental  Protection Agency;
November 1991.

     112U.S. EPA,  "Issues Relating to VOC  Regulation Cutpoints,
Deficiencies, and Deviations.   Clarification to Appendix D of
November 24, 1987 FEDERAL REGISTER," Office of Air Quality

                               374

-------
control strategies have small costs or even result in savings.
However, many other VOC control options have significant costs
associated with them.  For example, the estimated cost-
effectiveness of reducing emissions from automobile and light
truck coating operations in assembly plants is $1,000-4,000/ton
VOC.  Reducing emissions from the production of pneumatic rubber
tires is estimated to cost between $150 and $18,800 per ton of
VOC reduced.  The cost-effectiveness depends on the operation to
which control is applied.  Control of emissions from floating
roof tanks used for storage of petroleum liquids can cost up to
$3,700/ton VOC reduced.  Reducing emissions from the production
of high density polyethylene, polypropylene, and polystyrene
resins can cost between $1,000 and $3,000/ton VOC reduced
depending on the level of control required at these facilities.

     The comparative cost-effectiveness of stationary source NOx
emission controls was based on those suggested for utility
boilers.113  In ozone nonattainment areas,  standards are being
considered that will require controls more stringent than
suggested by reasonable achievable control technology  (RACT)
standards.  The RACT standards will likely be met through the use
of low NOx burner technology.  This technology has a relatively
low cost-effectiveness at up to $1,000/ton, but the achievable
emissions reduction is limited.  In order to meet the ozone air
quality standard in many areas, additional control will likely be
required particularly by the year 2000.  One of the likely
strategies utilized will be selective catalytic reduction (SCR)
which is estimated to cost $3,000-$10,000/ton NOx.

b.   Mobile Source Emission Control Cost-Effectiveness

     Although there is some merit to comparison with stationary
source control options, it is more appropriate to compare the
cost-effectiveness of Phase II RFG with other mobile source
control programs.  Emissions from mobile sources occur at ground-
level and are more widely dispersed than emissions from a large
point source.  Control of these emissions is therefore more
important in trying to reduce the formation of ozone, because
mobile source emission control (through fuel or vehicle measures)
will help all areas to reduce ozone, while controlling point
source emissions will only help a narrow area affected by the
emission plume from that source.   Hence, estimates of the cost-
Planning and Standards, May  25,  1988  (Revised 1/11/90).

     """""Evaluation and Costing of NOx Controls  for  Existing
Utility Boilers in the NESCAUM Region";  Draft Report prepared
by Acurex Corp., prepared  for Bill  Neuffer,  OAQPS,  U.S.  EPA,
October, 1992.

                               375

-------
effectiveness of various mobile source control strategies has
also been examined.

(1) .  Mobile Source VOC Control

     The cost-effectiveness of several VOC control strategies has
been identified for comparison with the Phase II RFG VOC
standard.  Enhanced inspection and maintenance (I/M)  programs
will cost between $900-1,700/ton VOC reduced, while basic I/M was
estimated to cost $5,400/ton VOC.114  The Tier 1 standards for
light duty vehicles (already implemented for the 1994 model year)
were estimated to cost about $6,000/ton VOC.  On-board
diagnostics are estimated to cost about $2,000 per ton of VOC
reduced as a result of this computerized engine monitoring
system.

(2).  Mobile Source NOx Control

      The enhanced I/M program is primarily a VOC and CO control
strategy, although NOx reductions are achieved.  The cost of NOx
emissions reduction is difficult to separate from the costs of
VOC and CO emissions reductions.  The cost-effectiveness of NOx
control through enhanced I/M programs is estimated to be $3,000-
6, 000/ton115.    An enhanced I/M program which employs  a more
stringent NOx cutpoint than typically assumed is estimated to
have a cost-effectiveness of $4,000-8,000/ton.

     The Tier 1 emissions standards for light duty vehicles will
incur an estimated incremental cost of $2,000-6,000/ton NOx if
credit is only given for those emission reductions achieved in
ozone nonattainment areas during the ozone season (to allow
direct comparison with reformulated gasoline).  Achieving the
Tier 2 mobile source NOx standards (should EPA determine that
such standards are necessary to meet air quality requirements) is
likely to cost more than $10,000/ton of NOx reduced.

     Certain NOx controls for heavy duty vehicles (both highway
and nonroad)  are likely to be very cost effective in comparison
to other programs.  Since heavy duty vehicles contribute a large
fraction of the NOx inventory, control of emissions from these
vehicles has the potential to result in large reductions.  EPA is
developing and studying such controls.  However,  heavy duty NOx
controls alone are unlikely to achieve the in-use reductions in
     114" Inspection/Maintenance Program Requirements,"  Final
Rule, November 5, 1992, 57  FR  52984.

     11540 CFR Part 51  (52950)  Inspection/Maintenance Program
Requirements; Final Rule; Environmental  Protection Agency,
November 5, 1992.

                               376

-------
NOx necessary to achieve the ozone NAAQS in most areas.  All NOx
control strategies that are not implemented or are not already
under consideration are likely to be very costly; NOx control
combined with the reformulated gasoline program is very
reasonable by contrast.

2.   Cost-Effectiveness Cutoff Limits

     In the NPRM, EPA proposed determining the Phase II RFG
standards based on a predetermined maximum limit for the cost-
effectiveness of that level of control.  Two cost-effectiveness
limits were presented, $5,000 and $10,000 per ton, and comments
were requested regarding the appropriate limit to select.
Several comments were received which addressed this subject.  One
commenter stated that the $5,000/ton VOC limit is appropriate.
Another stated that the $5,000-10,000/ton range proposed is the
upper end of costs incurred by many areas to achieve VOC control,
in effect establishing more costly criteria than required for
other programs.  The recommendation that EPA choose a low cost-
effectiveness target because of weaknesses in the analysis  (as
presented in the draft RIA) was also received.  However, EPA also
received the comment that the upper end of the proposed range was
reasonable considering the costs currently incurred by many
states for existing control programs.   (No support was given for
this last statement.)

     As discussed in the previous section, EPA has examined the
cost-effectiveness of a variety of stationary and mobile source
VOC and NOx control options.  Depending on the type of programs
and level of reduction achieved, these programs have a cost-
effectiveness ranging from essentially zero (or even a cost
savings)  to well over $10,000/ton reduced.  After considering
this information, EPA believes that a cost-effectiveness
benchmark of $10,000/ton is too high at this time and that a
benchmark around $5,000/ton is more appropriate for the Phase II
VOC standard and accompanying NOx standard.   However, as
explained elsewhere in this document and in the preamble for this
rulemaking, other factors were taken into consideration when
setting the standards, and hence, the cost-effectiveness limit
was not the only deciding factor.

3.   Other Economic Considerations

     The economic impact on the refining industry of the Phase II
reformulated gasoline standards and NOx standards for
reformulated gasoline will depend on the severity of the
standard.  It will also depend on the fuel parameter changes
which are necessary to achieve those standards.  The costs and
cost-effectiveness analyzed here are for the average refinery,
and thus, across the industry there will be a variety of costs
and economic impacts of this regulation.  The cost experienced by
the consumer, whether inside or outside of an ozone nonattainment

                               377

-------
area, will depend not only on manufacturing costs but also on
market conditions.

     While it is true that the cost-effectiveness for an
individual refinery may be different than that used here to set
the standards, EPA believes that the standards have been set with
enough flexibility to prevent refiners from incurring extreme
costs.  It is not currently possible to set the standards taking
into account the individual cost-effectiveness of each refinery
or the emissions reductions achieved in every distinct area using
RFC.  Such an analysis would require knowledge of the operating
costs and configuration of each refinery,  as well as the volume
of reformulated gasoline produced.  However, because EPA
considered the ability of refiners to achieve the emissions
reductions and thus did not specify a single fuel formulation to
meet the Phase II standards, each refiner will be able to choose
the refinery configuration that will produce RFC in the most
cost-effective manner for that refinery.  For example, if the
refiner produces a gasoline that is high in olefins, it is likely
that he can still meet the NOx standard without incurring costs
greater than $10,000/ton NOx (as the costs in Table VI-21 show.
Similarly, if a refiner produces a fuel that is high in
aromatics, he can meet the VOC standard without incurring
significantly higher costs.

4.    Energy Factors

     Production of Phase II reformulated gasoline subject to
performance standards for VOC,  NOx, and toxics will require an
increase in the amount of energy used at the refinery.  An
estimate of the energy used depends on many factors, including
how the energy balance is evaluated, the type and source of
oxygenate, the refinery configuration, and the reformulation
approach.  Determining an exact energy increase associated with
reformulated gasoline production  (on the basis of a constant
level of gasoline energy produced) is difficult.

     Cost-effective control of VOC and NOx will likely be
achieved largely through reductions in the sulfur content and
Reid vapor pressure (RVP) of the fuel.  Removing sulfur from
gasoline is an energy intensive process; the amount of energy
required varies depending on the sulfur level of the crude used
by the refinery.  Reducing the RVP of the fuel requires removal
of the lighter compounds in the fuel, also an energy consuming
process.  Overall, it is expected that the energy consumption by
refineries in producing Phase II reformulated gasoline will
increase slightly over the level of energy used to make Phase I
RFG, but the magnitude of this increase is difficult to measure
due to the many variables involved.

     As discussed in subsection C.4, changes in some parameters
impact the energy density of the resulting gasoline (and hence,

                               378

-------
the fuel economy achieved by the vehicle).   Although there may be
some increase in energy consumed in the form of gasoline as a
result of this (it could take more energy to go one mile),
especially as a result of the addition of oxygen, the impact for
Phase II fuels is likely to be small.  This is because few of the
fuel changes which would negatively impact fuel economy, such as
increasing the oxygen content to 2.7 wt%,  are likely to be cost-
effective for refiners to use in meeting the standards.

     There may also be energy benefits resulting from reduced
consumption of gasoline in areas using reformulated gasoline.
Historically, when the retail price of gasoline has risen,
consumption has declined.  Phase II reformulated gasoline is
expected to cost 1.2C/gallon more than Phase I RFC.  It is
possible that consumers may reduce gasoline consumption in
response to this price increase.  As a result, less petroleum
will be needed to produce gasoline,  potentially leading to a
reduction in oil imports.  However,  the reduction gasoline
consumption resulting from the increased price of Phase II RFC is
likely to be small compared to the likely impact of Phase I RFC.

5.    Environmental and Health Factors

     In response to the proposed Phase II RFC standards, EPA
received the comment that EPA should detail the value of the
health, welfare,  and environmental improvements achieved through
the standards.  Specifically, it was suggested that these
benefits be demonstrated in terms of the VOC and NOx reductions
achieved.  Section F presented EPA's analysis of the cost-
effectiveness of various levels of VOC and NOx control, expressed
in terms of the cost per ton of pollutant reduced.  The following
paragraphs discuss the health and welfare benefits of achieving
these reductions.

     The purpose of the reformulated gasoline program is to
reduce ozone formation and cancer incidences due to motor vehicle
emissions of VOC and toxics in those areas most in need of such
reductions.  The most widespread problems among the major air
pollutants for which an NAAQS standard has been established is
ozone.  Ozone is formed through atmospheric reactions of NOx and
hydrocarbons in the presence of sunlight.   Emissions from both
mobile and stationary sources contribute to ozone formation.

     Reducing ozone levels (through reductions in emissions of
VOC and/or NOx) in highly populated urban areas would help to
reduce short-term health effects caused by exposure periods as
short as one hour, such as impaired lung function, cough,  nausea,
chest pain, throat irritation, increased susceptibility to
respiratory infection, and increased sensitivity of asthmatics to
allergens  (e.g.,  pollen) and other bronchoconstrictors.  Exposure
to elevated ozone levels can even affect respiratory performance
in healthy individuals during exercise outdoors.  Long-term

                               379

-------
health effects of exposure to ozone include accelerated aging of
the lungs, reduces elasticity of the lungs, scarring of lung
tissue, and permanent reductions in baseline lung function.

     Reducing ozone levels in rural areas where fewer people are
exposed would enhance agricultural crop yield production,
currently estimated to be reduced by as much as 2 to 3 billion
dollars per year by existing ozone concentrations.116  In
addition, lower ozone levels would help reduce damage to forest
ecosystems which experience lower tree growth rate,  foliar
damage, and increased susceptibility to stress (e.g., insects,
disease,  drought) caused by current tropospheric ozone levels.
Ozone can also damage watershed areas and marine life.117  It
persists for several days in the lower atmosphere and can travel
hundreds of miles.

     The environmental benefits to the areas receiving
reformulated gasoline are measured in terms of tons of VOC
emissions reduced and number of cancer incidences avoided.  The
benefits of the NOx reduction program similarly are measured in
terms of the tons of NOx reduced.  In addition to these
measurable benefits in terms of ozone precursor reduction and
human health, there are many other health and environmental
benefits of reducing VOC, NOx,  and toxics emissions.  Additional
information on the benefits of NOx reductions can be found in
section B above.

     Reductions in mobile source emissions of air toxics
(benzene, 1,3-butadiene, formaldehyde, acetaldehyde, and POM) may
result in fewer cancer incidences. The number of cancer
incidences likely to be reduced as a result of the Phase II
standards has been estimated to be quite small, as summarized in
subsection F.3.

     A number of adverse non-cancer health effects have also been
associated with exposure to air toxics, particularly with higher
level exposures experienced in particular microenvironments such
as parking garages.  Benzene is known to cause disorders of the
blood.  It can also be immunotoxic, and may be a developmental
and reproductive toxicant.  High level exposure to 1,3-butadiene
can cause eye, nose, and throat irritation, while very high
exposure can cause effects on the brain leading to respiratory
paralysis and even death.  Epidemiologic studies of 1,3-butadiene
have also indicated a possible association with heart disease,
     116U.S. EPA, "Review of the National Air  Quality Standards
for Ozone  - Assessment of  Scientific  and Technical
Information,"  OAQPS, EPA-450/2-92-001, June  1989, p.X-25.

     117Ibid., p.X-26,27.

                               380

-------
blood disease, and lung disease at relatively high exposures.
Exposure to low airborne concentrations of formaldehyde can lead
to irritation of the mucus membranes of the eyes and respiratory
tract and potentially to cellular changes in the nasal cavity.118
A more thorough discussion of the variety of possible non-cancer
effects of concern from exposure to air toxics is contained in
EPA's Motor Vehicle-Related Air Toxics Study.119

     In addition to the ozone reductions and decrease in cancer
incidences which will be experienced as a result of the
reformulated gasoline program, other benefits are likely to
occur.  For example, the 2.0 volume percent oxygen requirement
for RFC will reduce emissions of carbon monoxide (CO) by 10-15
percent, an added benefit to areas which exceed the NAAQS
standard for that pollutant.  This will occur primarily in Phase
I of the RFC program, when the oxygenate mandate takes effect.
Additional reductions in CO emissions are likely to occur with
Phase II RFG, as well.  However, these Phase II benefits cannot
currently be quantified as a complex model for CO emissions has
not yet been developed.

     In areas not receiving reformulated gasoline,  no detrimental
effects are expected since the gasoline sold in those areas must
comply with the anti-dumping provisions.  Additionally, it is
likely that spillover of reformulated gasoline into these areas
will provide even cleaner fuel to those areas.

6.   Technological Feasibility

     The technological feasibility of producing fuels to meet the
Phase II standards was also considered.  EPA believes that the
refinery modelling results  (from which the fuel parameter control
costs were estimated) provide adequate support for the
feasibility of making the fuel parameter changes used to achieve
the standards.  The refinery models utilize only well-developed,
demonstrated, commercially available technologies,  and hence will
only produce fuels within the limits of these technologies.
Furthermore, one comment received on the proposal supported the
technical feasibility of Phase II RFG controls by noting that the
California Air Resources Board has set specifications for
California Phase II RFG which demonstrate the feasibility of
going beyond 25% VOC emissions reductions.  Given the incentives
     118Cotruvo, Joseph A. et al,  "Formaldehyde  Risk Assessment
Update  (1991)," Toxic Air Pollutants  from Mobile Sources:
Emissions and Health Effects,  Proceedings of an International
Specialty Conference, Air and  Waste Management  Association,
VIP-23, 1992.

     119EPA document 420-R-93-005 , April  1993.

                               381

-------
created by this rulemaking, in all likelihood new technologies
will be developed between now and the year 2000 which will reduce
the costs for certain types of fuel parameter changes.  Thus,  EPA
believes that the determination of fuel parameter control costs
using the results of the existing refinery models is conservative
and that the feasibility of producing such emission-reducing
fuels is justifiable.  This position was supported by many of the
comments received.

7.   Fuel Safety Factors and Driveability Concerns

     Comments concerning the safety of low RVP fuels in vehicle
fuel tanks were received which suggested that a minimum RVP level
be established and required for RFG.  One comment suggested that
producers will make RFG with an RVP in the "high 5 psi range" if
faced with a standard for an RVP of 6.5 psi.   Since no safety
studies are available for fuels with such low volatilities, the
commenter recommended setting 7.0 psi as the minimum for RFG.
Another comment questioned whether the study presented by EPA in
the Draft RIA did not adequately address the transitional months
(May and September) or the low temperatures and low RVPs that
could be experienced in areas on the fringe of nonattainment
areas.

     In EPA's evaluation of the safety of vehicle fuel tanks with
low RVP fuels, a thorough analysis of the safety of low RVP fuel
for the summer months (May through September) was performed.
This analysis showed that gasoline with an RVP above 5.0 should
not yield a combustible mixture in the fuel tank under any likely
summertime temperatures.  This analysis was based on the Oth and
10th percentile lowest temperatures experienced during these
months in the nonattainment areas.  The analysis further showed
that a fuel tank filled with gasoline at an RVP of 6.0 psi or
greater would not contain a flammable mixture.  Extending these
results to other fuel storage vessels, it is unlikely that
gasoline with a minimum RVP of 6.0 psi would lead to flammability
problems in pipelines or storage tanks.

     Some concerns were raised about driveability problems
arising from the use of low RVP fuels.  No testing was performed
by the Agency or other organizations to evaluate driveability at
low ambient summer temperatures.  However, EPA expects that
driveability of a 6.5 psi fuel in the summer should be similar to
current winter fuels, based on the actual vapor pressure of fuels
currently in production.  EPA discussed this finding with
representatives of both the oil and the automotive industries,
who did not indicate a concern with fuel RVPs as low as 6.5 psi.
However, they did express an uneasiness with lower RVPs.  Until
such time as data can be gathered to fully evaluate the
driveability impacts of low RVP fuels, it is not possible to
determine the true driveability impacts of fuels with RVPs lower
than 6.5 psi.  However,  given that most refiners will likely

                               382

-------
produce gasoline with an RVP around 6.7 psi to meet the Phase II
standards, EPA believes it to be appropriate to treat 6.5 psi as
a practical lower limit for RVP.


I.    Standards

     In the draft RIA,  EPA presented cost-effectiveness analyses
for the use of Phase II reformulated gasoline in each of the two
VOC control regions.  Based on these analyses, the Agency
proposed a range of possible standards for each VOC control
region.  The standards were proposed as a range instead of a
single number because EPA was requesting comment on what level of
cost-effectiveness  ($5,000 or $10,000/ton)  was reasonable to
require, and because EPA was considering a variety of minimum
levels for the Reid vapor pressure of Phase II RFC.  The
following sections summarize comments which were received on the
proposed standards and present the final standards derived based
on the cost-effectiveness analysis presented in this document.

1.    Comments on Proposed Standards

     Several comments were received on the proposed standards.
One comment suggested that since refiners are likely to reduce
parameters to levels lower than the mandated limits to ensure
compliance with the standards, the cost analysis should be based
on a marginal increase in the standard to determine the true
cost-effectiveness of the program.  Another comment stressed that
EPA distinguish carefully between the RFG standards and the
characteristics of the gasoline that will be produced to meet the
standards within the compliance margins.  The concern expressed
by both commenters was that EPA would set standards without
considering the fact that refiners may be operating near the
"knee" of the cost-effectiveness curve  (the point beyond which
the costs increase dramatically for a relatively small reduction
in emissions) at the levels of some parameters required to meet
those standards.  Because refiners would likely produce fuels
with even lower levels of the gasoline parameters to ensure
compliance, they may be pushed past the knee in the curve for
some of the parameters.

     The cost-effectiveness analysis presented here, on which the
standards for Phase II RFG will be based, is inherently an
average cost analysis,  since it is based on regional refinery
modelling data.  Hence, the standards derived from this analysis
will be averaging standards.  There is no need for refiners to
have a compliance margin in meeting the averaging standard.  The
difference between the stringency of the averaging standard and
the per gallon standard is intended to be roughly equivalent to
the compliance margin,  as the provisions of section VII of the
preamble demonstrate.  As determined in Reg-Neg, this difference
is 1.5% VOC.  As long as the per gallon minimum does not

                               383

-------
constrain the ability to meet the averaging standard, variations
between gallons and measurement error can be accommodated.
Refiners will be able to meet the averaging standard exactly,  and
thus there is no need to artificially adjust the cost-
effectiveness analysis to allow for compliance with the averaging
standard.  Furthermore,  since refiners can choose which standard
each gallon will meet after it has been produced, any measurement
error is in their favor, since they can apply any gallon to the
averaging standard.  Hence, EPA's standards are actually
conservative.

     Another recommendation suggested that the incremental cost-
effectiveness analysis begin at the statutory minimum (20%
reduction in VOC, no net increase in NOx),  and that the standards
be set at a point on the cost curve to avoid excessive
incremental cost-effectiveness.  Since the standards are based on
an incremental cost-effectiveness analysis the starting point for
the analysis is irrelevant.  If the standards were set based on
the total cost-effectiveness incremental to the Phase I
standards, the concern raised by this comment would have been
appropriate.

     It was also suggested that EPA define the VOC and toxics
performance requirements (standards) based on the emissions
reductions achieved by a fuel meeting pre-specified design value
property limits.  The design values suggested by the commenter
were based on the parameter values used in the draft RIA to
determine the proposed standards.  In effect, this would set a
fuel specification to ensure that a certain level of VOC control
would be achieved,  using the complex model to determine that
level of control.  EPA believes that it is inappropriate to set
the standards based on design values.  This would severely limit
the flexibility of individual refiners in meeting the standards.
In effect, this approach would set a fuel specification; the
Phase II standards are meant to be a performance standards that
can be met with any fuel formulation that is cost-effective to an
individual refiner.  Furthermore, EPA has revised its cost-
effectiveness analysis using the final complex model and updated
cost information, so the parameter values used to obtain the
proposed standards no longer necessarily are necessary or
required to meet the new standards.

     Another comment suggested that EPA set the Phase II
standards somewhat less stringent than California Phase II
reformulated gasoline, and that the federal standards include a
cap on olefins, aromatics,  and oxygen.  EPA has set the standards
based on the need for emissions control, the cost-effectiveness
of that control, and on the burden to refiners of meeting the
requirements.  Since the purpose of the complex model is to allow
refiners greater flexibility in meeting the performance standards
under both Phase I and Phase II, it would be inappropriate to set
caps on any fuel parameter for Phase II RFG.  The caps on some

                               384

-------
parameters which EPA has instituted only apply to fuels certified
with the simple model.  These caps were only intended to provide
a safeguard on the emission performance of RFC fuels until such
time as those parameters could be modelled.

     Two comments concerned the joint requirements of VOC and NOx
standards.  The first said that the desire to control both VOC
and NOx was in conflict because the fuel parameter changes that
reduce NOx tend to increase VOC (and vice versa).   EPA's revised
analysis,  using the final version of the complex model, does not
support this claim.   According to EPA's analysis, it is possible
to reduce both VOC and NOx.  Reductions in RVP,  the main source
of VOC control, yields small (minimal)  reduction in NOx.
Similarly, sulfur reductions to reduce NOx also yield minimal
reductions in VOC.  The second comment supported the requirement
of a NOx standard if it was consistent with the goals of
maximizing VOC and toxics reductions.  Toxics reductions have
already been shown to be cost-ineffective.  As the cost-
effectiveness analysis in section F demonstrates,  NOx control can
be achieved with maximum VOC control in a manner which is most
cost effective for both pollutants.

2.    Final VOC Standards and NOx Standards

     Based on the cost-effectiveness analysis and other
considerations presented above, EPA is setting VOC reduction
standards for Phase II reformulated gasoline and concurrent NOx
reduction standards for gasoline sold in areas participating in
the RFG program beginning in the year 2000.  These standards are
shown in Table VI-23 below.  The combination of fuel parameters
on which the standards are based is just one of many fuel
formulations which could be used to achieve the standards.  It
must be stressed that these standards are performance standards
which may be met by the refiner's choice of fuel parameter
controls;  they are not specifications for fuel composition.
                               385

-------
Table VI-23: VOC Standards for Phase II Reformulated Gasoline and
     NOx  Reduction  Standards  (Percent Reduction  in  Emissions)
Controlled Emission
VOC
NOx
- per gallon
- averaging
- minimum

- per gallon
- averaging
- minimum

VOC Control
Region 1
27. 51
29.0
25.0
5.5
6.8
3.0
VOC Control
Region 2
25.9
27.4
23 .4
5.5
6.8
3.0
     per gallon basis would be VOC: 17.2%; NOx: 5.3%
                                                         psi  on a
     The 5.5% per gallon NOx reduction standard has been
determined to be the level of reduction that can be cost
effectively achieved.  This is based on control of sulfur down to
138 ppm.  The VOC standard was determined based on the level of
VOC control that can be cost effectively achieved through RVP
reduction  (i.e., down to 6.7 psi)  as well as the VOC reductions
achieved by reducing sulfur to meet the NOx standard.  If EPA had
elected to set only a VOC standard and had pushed RVP down to the
practical limit (6.5 psi, as discussed previously), essentially
the same level of VOC reduction would have been achieved.
However, setting both the VOC and NOx standards based on an RVP
of 6.5 psi and sulfur at 138 ppm would severely restrict the
flexibility of refiners.  Furthermore, such strict standards
could pose an undue burden on some refiners with baseline levels
of certain parameters (e.g., olefins)  above the average.  Hence,
EPA set the VOC and NOx standards based on the emissions
reductions that could be achieved with a fuel containing 138 ppm
sulfur and an RVP of 6.7 psi.  As this analysis shows, EPA is
setting a NOx standard that refiners can meet in a cost effective
manner without jeopardizing their ability to meet the VOC
standard for Phase II reformulated gasoline.  The cost-
effectiveness of this level of NOx reduction, if credit is given
for the additional VOC reductions achieved at this level of NOx
control (on the order of $500/ton VOC reduced),  is about
$5,000/ton NOx reduced.

     As a result of these standards, VOC emissions will be
reduced by about 10,000 tons in VOC control region 1  (southern)
areas each summer and 32,000 tons in VOC control region 2
(northern) areas.   In addition,  southern areas will experience a
reduction of about 8,600 tons NOx and northern areas will
                               386

-------
experience a reduction of 13,800 tons NOx.  The emissions
reductions experienced in southern areas are smaller than
experienced in northern areas due to the fact that southern areas
are already required to use fuels with lower Reid vapor
pressures, and thus the emissions reduction benefits of RFC use
in these areas is smaller.

3.    Toxics Standard

     The statute sets the minimum Phase II standard for toxics
reduction at 25%, although EPA has the authority to reduce this
down to 20% "based on technological feasibility, considering
cost."120  There was general support in the comments on the NPRM
for the fact that the cost-effectiveness of toxics control beyond
a 20% reduction is questionable.  No substantive comments were
received opposing the option of setting the standard at the
minimum 20% reduction.  The cost-effectiveness analysis presented
here indicates that the technology implemented by refiners to
comply with the required VOC and NOx reductions will result in a
26% reduction in toxics annually on average.  Hence, most
refiners will clearly be able to meet a 25% toxics standard.  For
certain non-average refiners with high baseline levels of various
parameters, however, compliance with the VOC and NOx standards
may not automatically lead to compliance with a 25% toxics
standard.  For these refiners, additional reductions in toxics to
meet the standard would require expensive controls, putting them
at a competitive disadvantage relative to those refiners who
comply with the toxics standard "for free" based on their
compliance with the VOC and NOx standards.

     If marginal reductions in air toxics were cost effective,
EPA would be justified in requiring their control.  However, this
is not the case.  The additional toxics control required by the
subset of refiners described above results in very high cost per
cancer incidence avoided, in excess of $100 million/CI, well
beyond the $1-10 million/CI which the Agency believes to be
achievable through other programs.  In addition, benzene
reduction  (which, along with aromatics reduction, is one of the
two likely means of reducing toxics)  would impact only emissions
of benzene, not 1,3-butadiene, which has been shown to be of
greater cancer-causing risk to the public.  (The statutory
requirements of section 211(k) requires a focus on reductions in
mass emissions of air toxics, not on a reduction in cancer risk,
and therefore does not permit EPA to set the standard based on
relative cancer risk.)  Even though a 25% toxics standard is
technologically feasible, the circumstances discussed above raise
     120The toxics standard is a requirement  for an average
percent reduction over the entire  year,  not  solely in the
summer  (high ozone) season.

                               387

-------
questions about the increased cost to this subset of refiners of
implementing additional toxics reduction technology.

     Based on these concerns, EPA is setting the per gallon
toxics standard at the statutory minimum 20% (and the averaging
standard at 21.5%).  Despite the fact that the toxics standard is
reduced to 20%, EPA nevertheless anticipates that the average in-
use toxics performance will still be above 25%,  as discussed
above.  Total toxics emissions will be reduced by about 370 tons
in VOC control region 1 annually, and 630 tons in VOC control
region 2.   Approximately 3-4 cancer incidences will be avoided
annually nationwide as a result of this level of toxics
reduction.
                               388

-------
J.   Appendix VI-A:  Tables VI-AI to VI-A6
                               389

-------
Table VI-A1  Phase II RFG Refining Costs for Individual Parameter Changes
Parameter
and Range of
Parameter
Change
RVP:
7.79 to 7.33
7.85 to 7.37
7.64 to 7.30
RVP:
7.33 to 7.09
7.37 to 7.16
7.30 to 6 . 94
RVP:
7.09 to 6.5
7.16 to 6.5
6 . 94 to 6.5
Oxygen :
2.1 to 2.7
2.1 to 2.7
2.1 to 2.7
Region
i
2
3
1
2
3
1
2
3
1
2
3
Cost
(C/qal)
0.309
0.255
0.175
0.170
0.121
0.215
0.568
0.482
0.219
0.800
0.566
0.521
Capital
Charge
(MM$)
79.6
28.4
137.6
57.6
10.2
225.5
69.8
146 .5
302.5
(62.6)
(52.0)
(322.7)
Capital
Cost 15%
and 10%
ROI
«Vqai)
15%: 0.339
10%: 0.257
0.106
0.080
0.226
0.171
0.245
0.185
0.038
0.029
0.371
0.281
0.297
0.224
0.545
0.412
0.498
0.377
(0.267)
(0.202)
(0.194)
(0.147)
(0.531)
(0.402)
Adjusted
Cost @
10% ROI
«Vgai)
.226
0.229
0.120
0.110
0.112
0.125
0.495
0.349
0.098
0.865
0.613
0.650
Cost per
Unit
Parameter
Change
0.492
0.477
0.352
0.461
0.533
0.347
0.839
0.530
0.223
1.44
1.02
1.08
Units
for
Cost
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/psi
c/qal/%
c/qal/%
c/qal/%
Nation-
wide
Final
Adjusted
Cost
0.412
0.416
0.427
1.14
                                                 390

-------
Table VI-A2  Phase II RFG Refining Costs for Individual Parameter Changes
Parameter
and Range of
Parameter
Change
Sulfur:
337 to 250
453 to 250
441 to 250
Sulfur:
250 to 160
250 to 160
250 to 160
Sulfur:
160 to 100
160 to 100
160 to 100
Sulfur:
100 to 50
100 to 50
100 to 50
Region
i
2
3
1
2
3
1
2
3
1
2
3
Cost
0.156
0.098
0.514
0.721
0.101
0.756
0.789
_
0.756
1.30
_
0.733
Capital
Charge
(MM$)
61.7
75.2
137.1
56.1
11.32
579.3
(18.0)
_
66.0
91.9
_
(86.9)
Capital
Cost 10%
and 15%
ROI
(c/gal)
15%: 0.263
10%: 0.199
0.280
0.212
0.226
0.171
0.239
0.181
0.0421
0.032
0.953
(0.077)
(0.058)
_
0.109
0.082
0.392
0.297
_
(0.143)
(0.108)
Adjusted
Cost @
10% ROI
(c/gal)
0.0925
0.0298
0.459
0.662
0.091
0.524
0.807

0.730
1.21
_
0.768
Cost per
Unit
Parameter
Change
0.00106
0. 00014
0.00240
0.00736
0.0010
0.00582
0.0135

0.0122
0.0241

0. 0154
Units
for
Cost
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
c/gal/ppm
Nation-
wide
Final
Adjusted
Cost
.00203
0.00625
0.0125
0.0149
                                                 391

-------
Table VI-A3  Phase II RFG Refining Costs for Individual Parameter Changes
Parameter
and Range of
Parameter
Change
Aromatics :
29.65 to 28
28.34 to 28
30.54 to 28
Aromatics :
28 to 24
28 to 24
28 to 24
Aromatics :
24 to 20
24 to 20
24 to 20
Olef ins:
11 to 8
8 to 5
Region
i
2
3
1
2
3
1
2
3
5
5
Cost
0.313
0.014
0.285
1.011
0.237
1.23
1.44
0.809
1.43
0.3
3.0
Capital
Charge
(MM$)
33.0
18.8
541
(275.0)
147.8
104
(38.0)
93.5
(83.0)


Capital
Cost 15%
and 10%
ROI
(c/gal)
0.140
0.106
0.070
0.053
0.890
0.673
(1.17)
(0.885)
0.550
0.416
0.171
0.129
(0.162)
(0.123)
0.348
0.263
(0.137)
(0.103)
0.138
0.104
0.620
0.469
Adjusted
Cost @
10% ROI
(c/gal)
0.278
(0.003)
0.068
1.297
0.104
1.19
1.481
0.724
1.46
0.390
2.00
Cost per
Unit
Parameter
Change
0.169
(0.009)
0.0269
0.324
0.0259
0.298
0.370
0.181
0.365
0.130
0.668
Units
for
Cost
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
Nation-
wide
Final
Adjusted
Cost
0.0664
0.305
0.367
0.179
0.922
                                                 392

-------
Table VI-A4  Phase II RFG Refining Costs for Individual Parameter Changes
Parameter
and Range of
Parameter
Change
E200 +2, +3:
50.37 to 54
52.97 to 54
54.9 to 58
E200 +3:
54 to 57
54 to 57
58 to 61
E200 +3:
57 to 60
57 to 60
61 to 64
E300:
84 to 88
88 to 91
91 to 96
Region
i
2
3
1
2
3
1
2
3
3
3
3
Cost
1.27
0.104
0.181
2.91
0.827
0.955
8.56
1.22
1.39
0.83
1.32
1.65
Capital
Charge
(MM$)
(in)
0
298
299
414
24
1037
30
(34)
410
910
2705
Capital
Cost 15%
and 10%
ROI
(c/gal)
(0.473)
(0.358)
0
0.490
0.371
1.27
0.961
1.54
1.17
0.039
0.030
4 .42
3.34
0.112
0.085
(0.056)
(0.423)
0.146
0.110
0.325
0.246
0.965
0.730
Adjusted
Cost @
10% ROI
(c/gal)
1.39
0.105
0.06
2.60
0.452
0.946
7.49
1.20
1.40
1.00
1.70
2.79
Cost per
Unit
Parameter
Change
0.384
0.102
0.020
0.867
1.151
0.315
2.50
0.400
0.466
0.251
0.568
0.558
Units
for
Cost
c/gal/%
c/qal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
c/gal/%
Nation-
wide
Final
Adjusted
Cost
0.125
0.439
0.989
0.285
0.645
0.634
                                                 393

-------
Table VI-A5     Lester's spreadsheet for class B
                                    394

-------
Table VI-A5, cont'd
                                    395

-------
Table VI-A5, cont'd
                                    396

-------
Table VI-A6     Lester's spreadsheet for class C
                                    397

-------
Table VI-A6, cont'd
                                    398

-------
Table VI-A6, cont'd
                                    399

-------
Figure VI-Al
                                    400

-------
Figure VI-A2
                                    401

-------
Figure VI-A3
                                    402

-------
VII. Anti-Dumping Requirements for Conventional Gasoline--RIA

A.   Introduction

     Section 211(k)(8)  of the Act requires that the Administrator
promulgate regulations ensuring that the gasoline sold in areas
not participating in the reformulated gasoline program not be
more polluting than it was in 1990.  Such gasoline is known as
conventional gasoline.   To implement this requirement, EPA is
promulgating requirements known as the anti-dumping provisions
for conventional gasoline producers and importers.  The purpose
of the anti-dumping provisions is to ensure that fuel components
that cause harmful emissions and that are removed from or limited
in reformulated gasoline not be "dumped" into conventional
gasoline, and to likewise ensure that environmentally beneficial
fuel components not be used to make reformulated gasoline to the
detriment of conventional gasoline.

      Specifically,  the Act requires that average per gallon
emissions of specified pollutants from conventional gasoline use
must not deteriorate relative to emissions from 1990 gasoline, on
a refiner, blender or importer basis.  Compliance is measured by
comparing emissions of a refiner's121 post-1994 conventional
gasoline against its baseline gasoline.  An individual baseline
is developed for each refiner based on the quality of its 1990
gasoline, although under certain circumstances the individual
baseline is set at the statutory baseline.  These requirements
apply to all conventional gasoline producers whether or not
reformulated gasoline is also produced.

B.   Emission Requirements

     The Act specifies that EPA regulations ensure that emissions
of VOC,  CO, NOx and toxics due to conventional gasoline use not
increase over 1990 levels, on a refiner basis.  Additionally,
each of the four pollutants is to be considered separately,
except that increases in NOx emissions due to oxygenate use may
be offset by equivalent or greater reductions in the other
pollutants.

     In the July 9,  1991 proposal, EPA discussed how it may be
possible to determine,  on a seasonal or yearround basis,  that
emissions of one or more pollutants from conventional gasoline
use may or may not increase.  Even though emissions reductions
     121 For ease in discussion, the term  "refiner",  as  used in
this discussion of the anti-dumping program,  will hereafter
include refiners, blenders  and importers.   Where appropriate,
blenders and importers will be mentioned  specifically.

                               403

-------
which may be expected to occur on average may not occur for every
refiner, blender and importer, any environmental detriment is
likely to be minimal.  Thus,  it is appropriate not to require
monitoring of those emissions that are not expected,  on average,
to increase.  Reduced monitoring and compliance requirements will
also result in cost savings.

     For several reasons, VOC and CO emission increases are not
expected in conventional gasoline areas,  and thus EPA is not
regulating these two pollutants under the anti-dumping
provisions.  As discussed in the July 9,  1991 proposal, Phase II
volatility requirements and spillover of oxygenated and
reformulated gasoline are expected to contribute to VOC
reductions in conventional gasoline areas.  Since outside of the
high ozone season there are no VOC controls imposed on
reformulated gasoline,  there is no need to impose VOC control on
conventional gasoline to counteract potential dumping.  EPA also
does not expect CO emissions to increase, in or out of the high
ozone season, primarily due to the oxygenated fuels program.

     Toxics and NOx emissions, however,  could increase in
conventional gasoline areas as a result of reformulated gasoline
dumping.  The reformulated gasoline benzene cap of 1.0 volume
percent, as well as the reduction in aromatics due to the volume
and octane contributions of oxygenate, could result in increased
toxics emissions in conventional gasoline due to dumping of
aromatics and benzene122.   Because the oxygenate and benzene
requirements are yearround, toxics emissions due to conventional
gasoline use could also increase yearround.  Likewise, the net
effect of conventional gasoline fuel component changes (due to
the production of reformulated gasoline)  could result in
increased conventional gasoline NOx emissions.  Because of the
potential for toxics and NOx emissions to increase, EPA is
regulating emissions of these two pollutants under the anti-
dumping provisions.

1.   Emission Requirements Prior to January 1, 1998

     Prior to mandatory use of the complex model on January 1,
1998, the requirements of section 211(k)(8) of the Act will be
met by requiring that the annual average exhaust benzene
emissions of a refiner's conventional gasoline not exceed its
baseline exhaust benzene emissions.
     122 It is possible that no dumping of benzene  and aromatics
would occur.  Refiners could  choose  to modify refinery
operations to reduce production  of benzene  and aromatics rather
than divert these two fuel components to conventional gasoline.

                               404

-------
     The exhaust benzene emissions due to conventional gasoline
may be determined using the simple model.  Only the effects of
fuel benzene and fuel aromatic content on exhaust benzene are
included in this model.  Monitoring of only the benzene and
aromatic content of conventional gasoline is sufficient during
this period because, by the simple model, these are the only fuel
components which will be removed from reformulated gasoline which
affect toxic emissions.

     Per the negotiated agreement, anti-dumping compliance using
the simple model also requires that the annual average sulfur,
olefin and T90 values of a party's conventional gasoline not
exceed its baseline values of those parameters by more than 25
percent.  These limits will provide additional assurance that
conventional gasoline emissions will not rise prior to use of the
complex model.  EPA does not expect the levels of these
parameters in conventional gasoline to naturally increase due to
the reformulated gasoline program, since the simple model for
reformulated gasoline caps these three fuel parameters at
baseline levels and thus does not require their reduction.
However, the relaxed caps for conventional gasoline sulfur,
olefin and T90 will provide some processing flexibility for
refiners while not harming the environment.

     The exhaust benzene emissions due to conventional gasoline
may also be determined using the complex model.  Because all of
the parameters affecting exhaust benzene emissions are part of
the model (benzene, aromatics, RVP, sulfur, olefins,  E300, E200,
and oxygen)  there is no need for separate "caps" on fuel
parameters as associated with the simple model.

     While a refiner may choose to use either the simple model or
the complex model prior to January 1,  1998, it must use the same
model for determining reformulated and conventional gasoline
compliance,  if both fuels are produced.  Commenters suggested
that use of the complex model (for reformulated gasoline
compliance)  by one refinery should not require it at others, nor
require it to be used to determine anti-dumping compliance.
However, the anti-dumping program is inherently tied to the
reformulated gasoline program in that the specific model used to
certify reformulated gasoline will affect which fuel components
are likely to be dumped.  Requiring use of the same model for
reformulated gasoline and anti-dumping will ensure that the
effect of these components on conventional gasoline emissions
will be the same as on reformulated gasoline emissions.  Without
use of the same model, incentives would exist to shift dirty
components to conventional fuel areas using whichever model
predicts the lowest emissions increase due to those components.
If no reformulated gasoline is produced, the refiner may use
either model to comply with the anti-dumping requirements.
                               405

-------
     A refiner's baseline exhaust benzene emissions are
determined by evaluating the refiner's baseline fuel parameter
values in the model chosen by the refiner.  At the end of a
compliance period, the average fuel parameter values of a
refiner's conventional gasoline over that period are evaluated in
the same model used to determine the refiner's baseline
emissions.  The resulting emission values are then compared to
the baseline emission values to determine if the party is in or
out of compliance with the anti-dumping requirements.

2.   Emission Requirements Beginning on January 1, 1998

     When complex model use becomes mandatory on January 1, 1998,
the requirements of section 211(k)(8)  of the Act shall be met by
requiring that the exhaust toxic emissions and the NOx emissions
of a party's conventional gasoline,  as determined using the
complex model, not exceed that party's baseline exhaust toxic and
NOx emissions.

     The exhaust toxics emissions requirement under mandatory use
of the complex model includes all five pollutants defined in
section 211(k)(10)(C)  as toxics.   These are exhaust benzene,
formaldehyde, acetaldehyde, 1,3-butadiene and POM.  Benzene
emissions occur in both exhaust and nonexhaust emissions, and
accordingly, §211(k)(10)(c) does not limit the toxic air
pollutant benzene to exhaust benzene.   However, as stated,  EPA is
only promulgating anti-dumping regulations applicable to exhaust
benzene.  Nonexhaust benzene emissions will be effectively
controlled by the summertime volatility controls applicable to
conventional gasoline.123  The sum of the baseline exhaust
emissions of each of these toxics is the value that must not be
exceeded by the sum of the exhaust emissions of each toxic
pollutant due to a refiner's or importer's post-1994 annual
average conventional gasoline.

     In addition to the exhaust toxics requirement discussed
above, conventional gasoline NOx emissions are also controlled
when use of the complex model becomes mandatory.  However,  the
Act states that any NOx emissions increase in conventional
gasoline due to oxygenate use can be offset by VOC, CO and toxic
emission reductions.  The VOC and CO emissions reductions which
occur with oxygenate use are much greater than any potential NOx
     123 No credit can be taken nor penalties  received under the
anti-dumping program for nonexhaust  benzene reductions,  or
increases.  Nonexhaust benzene emissions  will decrease,
relative to 1990, due to RVP  reductions,  which are a VOC
reduction strategy already  considered under the anti-dumping
program as the reason for not explicitly  controlling VOC
emissions.

                               406

-------
increase, as discussed in the July 9,  1991 proposal.  To allow
for the determination of NOx emissions without the effect of
oxygenate use, EPA is allowing NOx baseline emissions
determination and subsequent compliance to be determined on a
nonoxygenate or oxygenate basis, as will be discussed in
paragraph C.4.e.

C.    Individual Baseline Determination

1.    Requirements for Refiners, Blenders and Importers

a.    Requirements for Producers of Gasoline and Gasoline
     Blendstocks

     Each U.S. refinery which converts crude oil and its
derivatives into gasoline and gasoline blendstocks,  and which was
in operation for at least 6 months in 1990, must determine an
individual baseline using Method 1 and/or Method 2 and/or Method
3-type data (these three data types are discussed in paragraph
C.2).   Such a refinery is not allowed to have the statutory
baseline  (further discussed in paragraph C.2.e)  as its individual
baseline.  EPA is requiring at least 6 months of operation in
order to ensure that the data is sufficiently representative of
the refinery's 1990 gasoline.  A refinery in operation for less
than 6 months in 1990 gets the anti-dumping statutory baseline as
its individual baseline.

     The requirement that such a refinery determine an individual
baseline rather than use the statutory baseline is to minimize
negative competitive and environmental impacts.   It is possible
that,  in 1990, half of the refineries produced gasoline which
results in higher emissions than the statutory baseline and half
produced gasoline which results in lower emissions.   If the
statutory baseline were required,  those with higher emissions
would be penalized because they would have to comply with more
stringent requirements relative to complying with their own
baseline.  Those with lower emissions would be able to produce
dirtier gasoline than they did in 1990, and actually could
utilize components which were "dumped" (to produce reformulated
gasoline).  If the statutory baseline was allowed, but not
required, again those with lower emissions would benefit, and
would be the only ones to use the statutory baseline because they
could produce dirtier gasoline.  Those with higher baseline
emissions would not be penalized,  but their post-1994 gasoline
would not be able to be any dirtier than their 1990 gasoline.  In
both cases, the competitive and environmental impacts are
adverse.  To minimize these results, EPA is requiring that each
refiner develop sufficient data, by estimating the quality of
their 1990 gasoline based on actual 1990 and post-1990 refinery
data,  thereby avoiding default to the statutory baseline.
                               407

-------
     While baseline parameters may be estimated from data on
gasoline shipment or production information,  baseline volume is
that volume of gasoline which actually left the refinery in 1990.
While this may be more than was produced in 1990 due to shipments
made from inventory, fuel parameters are not likely to differ
much from 1990 production.  Additionally,  the volume of gasoline
which was shipped is what actually contributed to emissions.
Likewise, gasoline exported in 1990 is not included in the
baseline determination since that gasoline did not contribute to
motor vehicle emissions in this country.  The baseline of an
entity which produced only gasoline blendstocks in 1990 (and did
not produce gasoline) will be the anti-dumping statutory baseline
if in the future such an entity produces gasoline.

b.   Requirements for Purchasers of Gasoline and/or Gasoline
     Blendstocks

     Per the April 16, 1992 proposal, EPA is requiring that the
individual baseline of a refiner who exclusively purchases
blendstocks and/or gasoline and mixes these purchased components
to form another gasoline be determined using Method 1-type data.
Refiners engaged in this type of refinery operation are commonly
called "blenders" or "terminal operators".  If a blender does not
have data on every batch of 1990 gasoline for each of the
required parameters, its baseline will be the anti-dumping
statutory baseline.

     EPA also proposed this requirement in the July 9, 1991
proposal, at which time it discussed its reasoning behind
allowing blenders to only use Method 1-type data.  Method 2 or 3-
type data, i.e., data based on 1990 or post-1990 blendstocks,
would not be appropriate for determining the quality of gasoline
produced at a blending facility in 1990, since blendstocks can
vary from week to week in both proportion, type and composition.
Thus, it is not feasible to determine the quality of a blender's
1990 gasoline by analyzing post-1990 changes in refinery
operations.  To allow such data would create an incentive for
blenders to discover or develop 1990 and post-1990 data on
selected blendstocks that would yield relatively high baseline
emissions.  Confirming the accuracy of this data would be very
difficult.  Therefore, use of any data other than that on 1990
gasoline composition would not yield a confident estimate of the
quality and emissions of the blender's 1990 production.

     Some commenters suggested that blenders be allowed the same
opportunities as refiners to use 1990 and post-1990 gasoline and
blendstock data.  Otherwise, they claimed, a blender may have to
"reformulate" its conventional gasoline, and would thus be
penalized for not measuring a fuel parameter that was not
required to be measured in 1990.  As stated,  EPA does not believe
that use of blendstock data or post-1990 gasoline or blendstock
                               408

-------
data would allow an accurate portrayal of a blender's 1990
production.

     One commenter suggested that blenders with insufficient 1990
data, or those not in business in 1990, should have a more
stringent baseline than the statutory baseline because they would
have an advantage over half of the U.S. refiners (i.e.,  those
cleaner than the statutory baseline).   The Act requires EPA to
set the statutory baseline as the individual baseline for those
with insufficient data from which to develop an individual
baseline.  The Act does not require any entity to default to a
more stringent baseline than either their individual baseline or
the statutory baseline, depending on data sufficiency, as
determined by EPA.  Since there is neither a clear advantage nor
disadvantage for use of the statutory baseline by blenders (and
no additional supporting data was provided by the commenter to
support its claim) it is not appropriate to impose additional
requirements or a more stringent baseline on blenders.

     EPA is not requiring a blender who blended only gasoline and
oxygenates in 1990 and who will continue to only blend gasolines
and oxygenates in the future (post-1994) to establish a baseline,
although it may.  No environmental harm is expected because the
gasoline will already have been accounted for in the refiner's
baseline and compliance calculations.   Additionally, the CO and
VOC reductions from oxygenate use should more than offset any NOx
increases due to oxygenate.

     Finally, because of the blendstock accounting requirements
specified in §80.101, refiners as described in this paragraph are
required to determine a baseline API gravity (which can be
converted to specific gravity),  if using Method 1-type data to
determine an individual baseline.  Because API gravity is
routinely determined for gasoline, and thus should exist if other
Method 1-type data exists, this requirement, although not
proposed, does not impose significant additional burden on such
refiners.

c.   Requirements for Importers of Gasoline and Gasoline
     Blendstocks

     In the July 9,  1991 proposal, EPA proposed to allow persons
importing refinery-finished gasoline to establish a baseline
using Method 1,  2 or 3-type data.  As proposed,  the foreign
refineries in which the gasoline was produced would have to
establish a baseline in the same way as domestic refineries.   EPA
also proposed that the importer could only use Method 1-type data
if the gasoline was not refinery-finished, i.e., if it was
blended outside of a foreign refinery.  If lacking sufficient
Method 1-type data,  the importer would use the statutory baseline
as its individual baseline.
                               409

-------
     In the April 16, 1992 proposal, EPA proposed to allow
importers to develop a baseline using only Method 1-type data,  or
lacking sufficient Method 1-type data,  utilize the statutory
baseline.  EPA also proposed to allow importers which were also
foreign refiners, and which imported at least 75 percent of the
1990 gasoline production of its foreign refinery into the U.S.  in
1990, to develop a baseline for the refinery in the same way as
domestic refiners.

     Although this differed from the earlier 1991 proposal, EPA
stated that it believed that artificially high baseline emissions
could result if importers were generally allowed to use Methods 2
and 3-type data given the inability to track the 1990 gasoline
imports of most importers.  EPA also stated that to allow
importers to develop high baseline emissions without the data
required by Method 1 would encourage gaming, e.g., high-emissions
producing gasoline could leave the U.S. (from a refiner with a
relatively low baseline) and come back in via an importer with a
high baseline.  This clearly would be dumping, and would be
environmentally and competitively detrimental.

     EPA also stated that it did not believe that, in general,
importers, particularly those who are not refiners, would have
adequate 1990 data on their imported gasoline from which to
establish an individual baseline.  If importers were allowed to
establish individual baselines based on 1991/2 data, these
baselines could be relatively high since there would be no way to
1)  prevent the importer from choosing high emission gasoline to
import in 1991/2 simply to get an advantageous baseline, and 2)
insure the 1991/2 gasoline was the same as 1990 gasoline
imported.  Additionally, EPA stated that use of a baseline
auditor to verify an importer's baseline would likely be
ineffective unless the importer was also a refiner which met the
75 percent criteria.

     Most commenters agreed with EPA's proposal.  Some felt,
however, that the "75 percent" criteria was self-selecting--only
those importer/refiners with higher baseline emissions relative
to the statutory baseline would choose to develop an individual
baseline using Methods 1, 2 and/or 3.  Those importer/refiners
with relatively low baseline emissions would use the statutory
baseline.  Use of the statutory baseline could lead to dumping
since they would be complying with a baseline which was less
stringent than one based on their own 1990 gasoline quality.  EPA
agrees that "dumping" could occur, but expects it to be minimal
since few foreign refineries are likely to meet the "75 percent"
criterion.  However, requiring those that meet the "75 percent"
criterion to develop a baseline using Method 1-, 2- and 3-type
data is consistent with the requirements for domestic refiners,
which clearly meet the criterion  (i.e., in most, if not all
cases, more than 75 percent of domestic refineries' 1990 gasoline
was utilized in the U.S. in 1990) .

                               410

-------
     One commenter claimed that location, not percent of
production imported, dictates enforceability.   Enforcement of a
non-domestic refinery is governed less by location and more by
the willingness of the company and/or country to open its
refinery for compliance visitations.  Again, not many refineries
are expected to meet the 75 percent criteria,  and ones that
choose to use it to establish a baseline are likely to be
receptive to required visitations.  Another commenter
specifically stated that Canadian refineries should be treated
the same as domestic refineries for the purpose of establishing
baselines.  As stated, EPA believes that it will be relatively
easy to accurately determine the quality of the gasoline produced
in 1990 for sale in the U.S. at a refinery outside of the U.S. if
a significant amount of the production of the refinery came to
the U.S.  Independent of where the refinery is located,  if less
than this amount was imported, it will be more difficult to
combine information on refinery operations and blendstock and
gasoline data (i.e., Methods 2 and 3-type data) and allocate such
information so as to establish the quality of the refinery's 1990
gasoline which was sent to the U.S.  In any case, 1990 blendstock
data or post-1990 blendstock or gasoline data will not provide a
reasonable basis for estimating the qualities of gasoline
imported into the U.S. in 1990.

     Others felt that an importer should be allowed to use all
available data to establish a baseline and have its baseline
verified by an auditor.  However, as stated, EPA believes that
significant dumping could occur if all available data is allowed.
One commenter suggested that importers with insufficient 1990
data,  or those not in business in 1990, should have a more
stringent baseline than the statutory baseline because they would
have an advantage over half of the U.S. refiners  (those cleaner
than the statutory baseline).   As discussed previously for
blenders, however, the Act does not require those with
insufficient data to default to a more stringent baseline.
Another commenter stated that the statutory baseline composition
did not represent the quality of imported gasoline, and suggested
that importers should get a statutory baseline that better
represented the quality of such gasoline.  Again, Congress has
already established the statutory baseline that is applicable
when an importer has insufficient data on the gasoline it
imported into the U.S. in 1990.  While EPA has discretion in
determining the terms of the wintertime gasoline for purposes of
the statutory baseline, EPA is directed to establish the
specifications of "1990 industry average gasoline", and not some
subset of the industry average.

     EPA is retaining the requirements for importers essentially
as proposed on April 16, 1992.  Thus an importer who imported
gasoline into the U.S. in 1990 shall have its individual baseline
determined using Method 1-type data.  If the importer does not
have data for every batch of 1990 gasoline imported, for each of

                               411

-------
the required fuel parameters, it shall have the anti-dumping
statutory baseline as its individual baseline.  Additionally, An
importer who did not import gasoline  (e.g., an importer which
imported gasoline blendstocks or which did not import anything)
into the U.S. in 1990 but who does so after 1994,  shall have the
anti-dumping statutory baseline as its individual baseline.
Finally, an importer which is also a refiner that imported more
than 75 percent of the 1990 gasoline production of a refinery
into the U.S. in 1990, must determine a baseline for that
refinery in the same manner as domestic refiners,  i.e., using
Method 1-,  2- and or 3-type data.

2.    Types of Data

     Compliance under section 211(k)(8) of the Act is measured
against the individual baseline of each refiner.  As specified in
today's rulemaking, the baseline consists of the estimated 1990
emissions and/or appropriate fuel parameters and volumes of each
refiner.  Additionally,  the Act states that if no adequate or
reliable data exist regarding the gasoline sold by a refiner in
1990,  the refiner must use the statutory baseline gasoline fuel
parameters as its baseline fuel parameters.

     After consulting with refiners, EPA believes that there
likely will be insufficient directly measured 1990 fuel parameter
data available, even in the case of some of the largest refiners,
from which to determine individual baseline values for each of
the required fuel parameters.  At the same time, EPA is concerned
about the use of the statutory baseline parameters by those
without individual baseline data, as discussed earlier in
paragraph l.a.  For these reasons,  refiners are required to
estimate the quality of their 1990 gasoline based on actual 1990
and post-1990 refinery data.

     In the proposals, three methods  (Methods 1, 2 and 3) were
described for refiners to use to determine their baseline
parameter values.  The methods differed in the type of data used
in each.  In the proposals, a parameter value was to be
determined solely by a single method.   For example, if not enough
Method 1 data existed to determine a baseline aromatics value,
the baseline aromatics value would be determined using Method 2,
if enough Method 2 data existed.  However, requiring each
parameter to be determined only using one method could result in
some of the available data not being utilized.  In the above
example, the Method 1 data that did exist would not be used in
the baseline aromatics determination.   However, since Method 1-
type data is based on 1990 gasoline, it is the best, most
accurate data for determining a baseline value, and should be
used to the fullest extent possible.  Thus in order to improve
the use of available data so as to develop an accurate and
representative 1990 baseline, EPA is allowing baseline parameter
values to be determined using a combination of the methods, or

                               412

-------
data types, if necessary.  The three data types (corresponding to
the three proposed methods)  are described below.

     EPA considers the data required for any of the methods to be
adequate and reliable for the purposes of determining baselines,
The available data must be used in a hierarchical manner,
however, in that available Method 1-type data must be used first,
followed by available Method 2-type data, etc.  In other words,
if insufficient Method 1-type data exists from which to determine
a given parameter, Method 2 type data should be used to
supplement any available Method 1-type data.  If insufficient
Method 2-type data exists, Method 3-type data should be used to
supplement any available Method 1 or 2-type data.   In this way,
the best available data is utilized.  The data used in each
method should support the data and results in the other methods.
For instance, the sum of the gasoline blendstock volumes
determined or used in Methods 2 and 3 should closely approximate
the total gasoline production as determined in Method 1.

a.   Method 1

     Method 1-type data consists of a refiner's records of 1990
batches, or shipments if not batch-blended, of gasoline.
Specifically, the measured fuel parameter value and volume of
batches or shipments of gasoline produced or shipped but not
exported are included in the baseline determination,  as long as
no data is double-counted.  The distinction between a refinery's
1990 production of gasoline and its 1990 shipments of gasoline
was not discussed in the proposals.  However, because it is
likely that there were differences between 1990 production and
1990 shipments (e.g., due to shipping from inventory) available
data on gasoline produced or shipped in 1990 must be used in the
baseline determination (although, as stated, double-counting of
data must be avoided).   The phrase "batches, or shipments if not
batch blended" is a slight modification of the proposal language
which included only the word "shipments".  Commenters stated that
many shipments can be made from a single batch, and that sampling
the batch and shipments from that batch would result in double-
counting the contribution of that batch in the determination of a
fuel parameter value.  In order to avoid such double-counting,
EPA modified the language as shown.

     Method 1-type data is the most reliable data available for
determining baseline parameter values, as it is actual 1990
gasoline data.  However,  only those parameters which are subject
to pipeline specifications and which are relevant to the
reformulated gasoline program, e.g., RVP and distillation
properties, are likely to be determinable using Method 1-type
data.  Even if parameters were subject to pipeline
specifications, in some cases the specification was only a
threshold value.   For instance, one pipeline specification
requires gasoline to have no more than 5 volume percent benzene.

                               413

-------
Refiners did not likely test for the actual benzene content,
other than to determine or estimate that it was less than 5
volume percent.  Meeting such a threshold specification would not
provide enough data from which to determine a baseline parameter
value.  Most refiners did not measure the other fuel parameters
of interest (to the reformulated gasoline program)  which are not
pipeline specifications, because there were no such requirements.

     Although not specified in the proposals, EPA is requiring
that gasoline blendstocks which become gasoline solely upon the
addition of a specific type and amount of oxygenate be included
in the baseline determination.  Additionally, unless evidence is
provided which indicates that such blendstocks were blended with
less than 10 percent ethanol or nonethanol oxygenates, or were
not further modified downstream, the refiner shall assume that
said blendstocks were blended with ten (10.0) volume percent
ethanol.  This requirement provides some assurance that baseline
emissions are not artificially low due to selective inclusion or
exclusion of such blendstock.  Requiring that the blendstock be
assumed to have been blended with ethanol (unless otherwise
shown) will result in a more stringent baseline than if it were
assumed blended with a  (lower volume of a) different oxygenate,
or even not further modified.  Hence, the burden of proof of
actual disposition of such product is on the refiner.

     Additionally, gasoline entering a refinery by purchase or
transfer, including intracompany transfers,  and which leaves the
refinery unchanged from its arrival state, shall not be included
in the receiving refinery's baseline, since it will have already
been accounted for in the baseline of the original producer.

b.   Method 2

     Method 2-type data consists of a refiner's 1990 gasoline
blendstock composition data and 1990 production records.  Because
it consists of 1990 data, Method 2-type data is also expected to
be very reliable and useful in accurately estimating 1990
baseline parameter values.  Only data on those blendstocks used
in the production of gasoline within the refinery are used in the
baseline determination.  Most of the blendstocks used in a
refiner's gasoline are produced in the refinery, however, some
are purchased or otherwise transferred from outside of the
refinery.  All data on any blendstock used in the refinery's
gasoline, whether produced inside or purchased outside of the
refinery, is to be used in the baseline determination.

     The average fuel parameter value of each gasoline blendstock
is determined from the measured fuel parameter values and
associated volumes of each type of blendstock.  The associated
volume is the volume of a batch of blendstock, or,  for a
continuous process, it is the volume proportional to the amount
of the blendstock blended into gasoline in that month.

                               414

-------
     Some commenters suggested that accounting for blendstocks
which do not contribute to toxics emissions makes no sense.
However, the quality of all blendstocks used in 1990 gasoline
contributes to a refiner's baseline toxic emissions
determination, and some blendstocks tend to increase toxics while
other blendstocks decrease toxics.  Another commenter suggested
that blendstocks sold for off-site blending be included in the
baseline.  Such blendstocks are not included in a baseline
determination--only those blendstocks which contributed to a
refiner's 1990 gasoline production are included.   Some commenters
suggested that only streams that constituted a minimum fraction
of the gasoline, say 2 or 5 percent, be required to be included.
However, the quality of even small volume streams could
significantly affect baseline parameter and associated emissions,
and thus must be included.

c.    Method 3

     Method 3-type data consists of a refiner's post-1990
blendstock composition data and 1990 production records.   As will
be discussed, in some cases post-1990 gasoline data may also be
used in the determination of baseline fuel parameter values.  As
for Method 2-type data, when blendstock data is used, only
blendstocks used in the production of gasoline within the
refinery are evaluated for baseline purposes,  whether the
blendstocks were produced in the refinery or purchased or
transferred from an outside source.  The average fuel parameter
value of each gasoline blendstock is determined from the measured
post-1990 fuel parameter values and associated volumes of that
blendstock.  The average post-1990 fuel parameter value and the
total 1990 volume of each type of blendstock are then used in the
estimation of the baseline value of the parameter.

     In order to ensure that the fuel parameter values obtained
with Method 3-type data adequately represent the 1990 values of
those parameters, the refiner must provide a detailed
documentation and comparison of its 1990 and post-1990 refinery
operations, especially in areas which affect the fuel
parameter(s) being determined from Method 3-type data.
Circumstances which would cause its post-1990 gasoline to differ
from its 1990 gasoline must be accounted for in determining a
fuel parameter using Method 3-type data.  For instance, if post-
1991 data is used, appropriate adjustments must be made for the
refinery operational changes that occurred due to the 1992
volatility rules and the oxygenated fuels program, two situations
which could cause post-1990 operations to differ from 1990
operations.  The refiner must adjust its post-1990 data to
account for such situations, perhaps using refinery modeling.  In
effect, the refiner is "backing out" its 1990 parameter values
from its post-1990 data.  The required documentation will assist
the baseline auditor in its verification and EPA in its review of
the refiner's baseline submission.

                               415

-------
     In the case where a blendstock was used in a refiner's 1990
gasoline but is no longer used, the adjustments made to account
for 1990 and post-1990 operational differences will allow
determination of that blendstock's composition.  In the case
where a blendstock was not used in a refiner's 1990 gasoline,  but
is used now, no data on that blendstock is included in the
baseline determination.

     EPA received many comments stating that the use of post-1990
gasoline data was more accurate than using post-1990 blendstock
data for a Method 3-based determination.  In this regard, some
commenters suggested letting the baseline auditor verify that
differences between 1990 and post-1990 operations were accounted
for and that post-1990 gasoline represented 1990 gasoline.  In
April 1992, EPA proposed that post-1990 gasoline data could be
used if each blendstock in post-1990 gasoline was within 10
percent of the volumetric fraction of that blendstock in 1990
gasoline.  One commenter said the blendstock fractional
requirements were arbitrary, and that there should be no such
requirements--the auditor would verify the similarity between
1990 and post-1990 gasoline composition.  Another commenter
suggested that 10 percent was still too narrow, and that the
criterion should be 50 percent, or 10 percent if blendstock
fractions were determined relative to the volume of the total
blend,  not of only the blendstock.  Additionally, some commenters
cited the lower cost of collecting and analyzing gasoline rather
than blendstock data as a reason for allowing post-1990 gasoline
data to be used in the baseline determination.

     EPA is allowing post-1990 gasoline data to be used to
estimate 1990 baseline parameters, subject to certain
requirements.  In addition to requiring the same detailed
documentation of 1990 and post-1990 operations as required when
using post-1990 blendstock data, the volumetric fraction of each
blendstock in post-1990 gasoline must be within plus or minus ten
(10.0)  percent of the volumetric fraction of the same blendstock
in 1990 gasoline or be within plus or minus two  (2.0)  percent on
an absolute basis relative to the 1990 value.

     Under the first criterion, for example, if a refiner's 1990
gasoline contained 30 volume percent reformate, post-1990
gasoline data may be used (provided all other parameter values
also conformed to these requirements) as long as it contained
27.0-33.0 volume percent reformate, i.e., +/- 10 percent.  Under
the second criterion, for example, if a refiner's 1990 gasoline
contained 3.0 volume percent dimate, post-1990 gasoline may be
used (provided all other parameter values also conformed to these
requirements) as long as it contained 1.0-5.0 volume percent
dimate, i.e., +/- 2.0 percent on an absolute basis.

     These two criteria are expected to cover typical year-to-
year blendstock volumetric fraction fluctuations.  Larger

                               416

-------
differences would indicate more significant refinery operational
changes.  Post-1990 gasoline data for which a single 1990
blendstock does not meet either of the blendstock fraction
requirements cannot be used in the baseline determination.
However, EPA also received comment that many refiners would not
be able to use post-1990 gasoline data, even with the expanded
criteria,  simply due to butane utilization changes from 1990.
Because butane, and thus RVP,  were reduced after 1990 due to
volatility controls, and because RVP reductions reduce emissions,
EPA is exempting butane from the blendstock requirements for
using post-1990 gasoline.

     If post-1990 gasoline data does not meet the blendstock
fraction requirements described above, it will be difficult for a
refiner to show similarity or account for differences between
1990 and post-1990 operations without looking into post-1990
operations at the blendstock level,  thus making any gasoline data
moot.  Although the criteria specified for being allowed to use
post-1990 gasoline data may disadvantage some refiners which have
collected post-1990 gasoline data but not blendstock data (and
thus require them to collect additional data after the final
rule),  EPA has indicated throughout its proposals that Method 3
was primarily based on blendstock data, with gasoline data
allowed as the exception.

     Finally, as for Method 1-type data (i.e., gasoline data),
EPA is requiring that post-1990 gasoline blendstocks which could
become gasoline solely upon the addition of a specific type and
amount of oxygenate be included in the baseline determination, to
provide some assurance that baseline emissions are not
artificially low due to selective inclusion or exclusion of such
blendstocks.

     While the data for Method 1 is obviously more reliable than
that required for Method 3, the Agency considers reliance on the
data for Method 3 appropriate because it 1) provides a fair
approximation of 1990 gasoline, and 2) provides an additional
means for determining baseline parameters,  thus not requiring the
statutory baseline for refiners with insufficient 1990 data for
Method 1.   As discussed, use of the statutory baseline for
refiners should be avoided because of potential and likely
negative environmental and competitive effects.

d.   E200 and E300

     Although not previously included among the fuel parameters
for which baseline values are required to be determined, EPA is
now requiring that baseline values be determined for the fuel
parameters E200 and E300, the percent evaporated at 200°F and
300°F,  respectively.  Although these two fuel parameters replace
T50 and T90,  respectively, in the complex model, baseline values
                               417

-------
of T90 are still required to be determined if producing
reformulated gasoline or using the simple model.

     EPA expects E200 and E300 values to be determined directly
from gasoline or blendstock data, even if distillation
information has to be regraphed.  If such a determination is not
possible, and  can be proven, E200 and E300 values may be
estimated from otherwise acceptable T50 and T90 data using the
following equations:

          E200 = 147.91 - (0.49 x T50)
          E300 = 155.47 - (0.22 x T90)


e.   Anti-Dumping Statutory Baseline and Emissions

     As mentioned earlier,  in some cases a blender or importer
may not be able, or be allowed, to develop an individual baseline
from its own data.  In that case, the blender or importer would
have the statutory baseline as its individual baseline.  Although
the compliance period for conventional gasoline is annual (as
discussed in the proposals and in §80.101 of this rulemaking),
emissions determined using the complex models are determined on a
summer and winter basis.  Thus, separate anti-dumping summer and
winter baseline fuel parameter values are required.  For those
without unique individual baselines, these values are the
statutory summer baseline specified in the Act, and the winter
baseline determined by EPA as required by the Act  (and as
specified in Section II of this document)124.

     Because of the blendstock accounting requirements specified
in §80.101,  a baseline statutory value API gravity (which can be
converted to specific gravity) is also required.  As indicated in
section 211(k)(10) of the Act, the summer value is 57.4 °API.
The corresponding winter value determined by EPA is 60.2 °API.

     Because separate summer and winter baseline fuel parameter
values are required, the single set of fuel parameters specified
in the proposals as the anti-dumping statutory baseline is
basically moot with regard to the final rule, except for annual
average values of sulfur, olefins and T90.  However,  for
clarification and completeness, it is discussed here.  Few
substantive comments were received concerning the proposed annual
average statutory baseline (which is a weighted average of the
summer and winter statutory values), and none led to a change in
the proposed fuel parameter values.   However, the baseline listed
     124 As discussed in Section  IV, when  the  complex model is
used to determine emissions, winter RVP is held at 8.7 psi
because there are no RVP effects on winter emissions.

                               418

-------
in the February 1993 proposal reflected a nonhigh ozone season of
September 16-March 31 (and thus a high ozone season of April 1-
September 15).   The current high ozone season (upstream from
retail outlets) is May 1-September 15.  The proposed anti-dumping
statutory baseline has been modified to reflect the change in
fuel parameter values and fuel consumption due to the change in
the length of the high ozone season,  and is shown below.
          Benzene, volume percent                 1.60
          Aromatics, volume percent        28.6
          Olefins, volume percent                 10.8
          RVP,  psi                                       8.7125
          T50,  degrees F                                 207
          T90,  degrees F                                 332
          E200, percent                                  46.5
          E300, percent                                  82.4
          Sulfur, ppm                             338
          API Gravity                             59.1

     Finally,  the annual average anti-dumping baseline emissions
are shown below.  These emissions were determined by evaluating
the summer and winter statutory baseline fuel parameters in the
appropriate summer or winter model or equation,  and then
determining annual average emissions by weighting the summer and
winter emissions by the fuel consumption fractions in the high
ozone and nonhigh ozone seasons, 39.6% and 60.4%, respectively.
As stated, annual average sulfur, olefins and T90 are shown
above.

                                                  Phase I
     Phase II
Exhaust Benzene--Simple Model                 6.45
Exhaust Benzene--Complex Model(mg/mile)       33.03
Exhaust Toxics (mg/mile)                             50.67
       104.5
NOx (mg/mile)                                       714.4
      1461.
3.   Data Collection and Testing Requirements

a.   Sampling Requirements
     125 The RVP that results from combining  the  CAA statutory
baseline RVP of 8.7 psi and the wintertime statutory baseline
RVP  (determined by EPA) of 11.5 psi  is  10.4  psi.  However,  as
discussed above, an RVP of 8.7 is used  in the winter complex
models.  Hence, the annual average RVP  shown here is 8.7 psi.

                               419

-------
     In the April 16,  1992 proposal,  EPA briefly discussed
sampling requirements for obtaining data from which to
determine baselines.  In the February 26, 1993 proposal, EPA
discussed additional requirements and changes to the
requirements previously discussed.  The primary reasons for
specifying sampling requirements is to ensure that enough
gasoline or blendstock samples were taken from which to develop
a representative baseline, and to minimize the picking and
choosing of data.

     In the April 16,  1992 proposal,  EPA stated that sufficient
sampling must be done, implying that the responsibility for
sampling was left to the refiner, but would have to be verified
by the auditor, and of course, ultimately accepted by EPA.  In
the February 26, 1993 proposal, EPA specified specific minimums
in order to ensure acquisition of adequate and representative
data.  Many commenters commented that they had sampled based on
the earlier proposal requirements for sufficient sampling, and
that EPA's later proposal could void data collected, and the
time and money spent.   EPA still believes it is the better
course to specify minimum sampling requirements, and thus is
specifying that for Method 1 and Method 3 gasoline data, at
least half of the batches (by number of batches, not volume),
or shipments if not batch blended, in a calendar month shall
have been tested for a particular parameter.  For Method 2 and
Method 3 blendstock data, at least weekly sampling of
continuous blendstock streams and, if blendstocks are produced
on a batch basis, sampling of at least half of the batches of
each blendstock produced in a month is required.

     However, EPA believes that at least some of the sampling
performed in response to the April 1992 proposal may well be
adequate.  To void such sampling would be an unnecessary waste
of resources if in fact it is adequate and representative of
the gasoline or blendstock produced or imported over a certain
period of time.  To avoid this result, EPA will also accept
less than the minimum data if the refiner can explain why the
data is less than the minimum requirements and why it is
sufficient in quantity and quality to use in the baseline
determination.  Although less than the minimum data required
may be used, all available samples must be analyzed and the
results used in baseline determination if more than the minimum
number of samples are available.  In addition, the auditor must
verify all of the above information and conclude that the
sampling is sufficient in quantity and quality to use in
baseline determination.

     Since 1990 data is the best data from which to calculate
baseline fuel parameter values, it makes sense that data
collected closer to 1990 is more easily adjusted to estimate
1990 values than data collected later.  Thus to prevent using

                               420

-------
more data than is necessary in the baseline calculation
(because excessive data may skew the results or allow gaming),
only data collected through the end of the calendar year in
which the minimum data requirements are met are allowed to be
used in the calculation of a given baseline parameter.
However, data collection may also cease when the minimum
requirements are met.  In any case, all data collected through
the time data ceased to be collected for baseline development
purposes must be included in the baseline determination.
Picking and choosing of data should thus be minimized even if
less than the minimum data is utilized because all data
collected must be included in the baseline determination126.

     At least three months of summer and three months of winter
data must be available for the determination of a parameter
value.  Summer months include any month in which any volume of
gasoline was produced to meet the federal summer volatility
requirements.  Winter is any month which could not be
considered a summer month.  Thus transition months may be used
to satisfy the three months of summer data requirement.  This
requirement will insure that the average parameter values
calculated by the above methods will adequately represent a
refiner's 1990 gasoline quality and composition, including
representative summer and winter data.  One commenter suggested
that if annual data were available, fuel parameters should be
allowed to be calculated on an annual basis instead of
requiring separate summer/winter determinations.  EPA will
accept annual data, however, because annual average baseline
emissions are determined on a seasonal basis first, separate
summer and winter average baseline fuel parameter values must
be determined from the annual data.  Some commenters suggested
that EPA accept less than three months of summer and winter
data if it could be shown that the composition of a blendstock
stream did not change seasonally.  Although in certain cases
this may be true, it is more efficient to require all refiners
to meet the minimum requirements than to ascertain whether the
quality of a few streams of a few refiners did not change
seasonally.  In certain instances where a refiner markets its
product only in an area or areas which experience no seasonal
changes relative to gasoline requirements, e.g., Hawaii, only
three months of data are required.
     126 Certain data may be excluded from the baseline
determination, if it can be shown that the sample was
mislabeled or incorrectly analyzed or if the results are
significantly different than would be expected based on
operating conditions, etc.  Such data must be reported, and the
auditor must verify the reasoning in support of such an
exclusion.

                               421

-------
b.   Post-Final Rule Data Collection

     Many refiners began collecting data to be used in the
baseline determination soon after the proposals were published.
They were of course taking a risk that significant changes
could occur between the proposals and the final rule.
Nonetheless, many have estimated their baseline parameter and
emissions values, and are working with potential auditors to
verify those determinations.  However, some refiners began
collecting data late, were waiting for the final rule or just
recently realized that they need additional data.  These
refiners are being allowed extra time to complete and submit
their baselines, however proof must be given that additional
data was needed and indeed was  collected after today.  The
requirement to collect at least three months of data on a
refiner's winter gasoline and at least three months of data on
a refiner's summer gasoline must still be met.

c.   Negligible Parameter Values

     In February 1993, EPA proposed that if a refiner can show,
and the auditor verify, that a refinery stream contains
negligible amounts of one or more of certain fuel properties
for which a baseline value must be determined, testing of that
stream for the fuel property(ies) is not necessary.  The fuel
properties discussed were aromatics, benzene, olefins,
saturates and sulfur.  Saturates are now excluded since a
baseline saturates value is not required to be determined.
Oxygen was not included in the proposal.  It is, however,
reasonable to include it because it is a required baseline
parameter, and is likely to exist in negligible amounts in many
blendstock streams.  Although each fuel component uniquely
affects emissions, the effect of assuming a zero value,
relative to "negligible" threshold values, is not expected to
significantly affect emissions.

     The proposed minimum values for a fuel component to be
negligible in a stream were 1.0 volume percent for benzene,
aromatics, olefins and saturates, and 10 ppm for sulfur.  The
final minimum values are shown below.

     Aromatics                1. 0 volume  percent
     Benzene                  0.15  volume percent
     Olefins                  1. 0 volume  percent
     Oxygen                   0.2 weight  percent
     Sulfur                   30.0  ppm

     Although not explicitly stated in the proposals, this
provision applies to blendstock streams on an average basis.
Thus fuel parameters deemed negligible for a given blendstock
stream are expected, on average, not to exist in the stream at

                               422

-------
greater than the appropriate minimum.  These final values
differ from those proposed based on commenter suggestions that
more appropriate negligible values should be approximately 10
percent of the expected value of the fuel parameter.  EPA
agreed with this for all of the parameters except for
aromatics, and modified the "negligible" threshold values for
these parameters.  The "10 percent" level for aromatics would
be about 2.8 volume percent.  Since some large streams have
aromatic contents of approximately this magnitude, it would not
be appropriate to credit a large stream with a zero value, even
though this would result in a more stringent baseline.  Thus
the negligible aromatics threshold is unchanged from the
proposal.

     As stated, a negligible oxygen threshold was not originally
proposed,  and in fact, there is no statutory baseline oxygen
value from which to estimate a 10 percent value.  Most non-
oxygenate refinery streams contain little or no oxygen.
Although accounting for small quantities of oxygen would reduce
baseline emissions, resulting in a more stringent baseline, the
minor environmental effects of having a slightly less stringent
baseline must be balanced against the costs of testing for such
small quantities.  Thus a de minimis level of 0.2 weight
percent was set as the negligible oxygen threshold.

     Any properties shown to exist in a refinery stream in
negligible amounts shall be assigned a value of 0.0.  One
commenter suggested accepting benzene, sulfur, olefins and
aromatics estimates of zero (for certain identified streams)
since any error would result in lower baseline emissions.
However, the fuel component values of a given stream do not
necessarily uniformly meet the negligible threshold values
described above.  Hence EPA is requiring each refiner to
determine individually for which streams certain fuel
components exist, on average,  in negligible amounts.

d.   Sample Compositing

     There have been no major changes in this provision from the
proposals.  Thus, samples that have been retained but not yet
analyzed may be mixed together in volumes proportional to the
volume of the batch, or shipment if not batch-blended, from
which the sample was taken and the mixture analyzed for the
required fuel parameters.  Blendstock samples obtained from
continuous processes over a calendar month may be mixed
together in equal volumes to form one blendstock sample  (e.g.,
all samples of reformate taken in a calendar month may be
mixed) and the sample analyzed for the required fuel
parameters.  Based on comments received, some of the details of
the compositing procedure have changed.  Specifically,
comparison, and adjustment, if necessary, of sample RVP is

                               423

-------
required, so as to avoid misrepresentation of a sample which
may have lost volatile components.  Additionally,  certain
compositing requirements were changed to better reflect
refinery conditions.

e.   Test Methods

     Many commenters were concerned that the test  methods they
had used to analyze samples would be invalid because they were
not the same as the required test methods being promulgated
today for reformulated gasoline.  EPA had proposed,  on April
16, 1992, that sampling and measurement techniques used to
determine baseline parameters must yield results which are
equivalent to the results obtained per the techniques and
methodologies specified for the reformulated gasoline program.
However, because of constantly evolving test methods, in
addition to the fact that the final regulations concerning
reformulated gasoline test methods will only be known today, it
would be inappropriate to disallow data not determined using
the test methodologies required in the reformulated gasoline
program.

     Nonetheless, EPA is concerned that the test methods used be
adequate.  In a modification of the proposal, EPA will accept
data determined using methods other than those required under
the reformulated gasoline program, upon petition and approval,
as long as the methodology or technique was a standard
industry-accepted measurement technique at the time the
measurement was taken.  The baseline auditor will  verify that
the techniques used to determine the baseline data meet either
of the requirements discussed above.

     It is possible that the industry-accepted methods in place
at the time will yield results which are less accurate than
more recent test methods or test methods required beginning in
1995.  EPA recognizes this fact, however, it would not be
appropriate to waste the resources expended by not accepting
data determined under earlier, industry-standard methods.
Also, no significant environmental deterioration,  if any, is
expected by accepting such data because development of new test
methods inherently involves consideration of previous test
methods, hence any differences should be minimal.   No
allowances will be made, however, for different analytical
methods utilized in the baseline and compliance calculations.
It is reasonable not to adjust baseline or compliance values
due to different measurement techniques because all refiners
should be affected in the same manner if industry standard test
methods were used to determine baseline parameter  values.
However, if data were determined under two test methods, EPA
may accept the data determined under the more accurate test
method, upon petition and approval.

                               424

-------
     Although not previously discussed,  oxygen content may be
estimated using oxygenate purchase, sale and/or blending
records or analytically.  However, the other fuel parameter
values of the oxygenate (e.g., the sulfur content of the
oxygenate) must be determined analytically,  or be deemed
negligible, if allowed.  All oxygen associated with allowable
gasoline oxygenates shall be included in the baseline oxygen
content if oxygen content is determined analytically.  Oxygen
content shall be assumed to be contributed solely by the
oxygenate indicated on the blending records, etc., if oxygen
content is determined from such records.

4 .    Baseline Fuel Parameter Determination

     The methodology for calculating baseline fuel parameters,
and changes in the methodology from the proposals, is discussed
below.  The equations and other details can be found at §80.91
in the regulations.
a.   Closely Integrated Gasoline Producing Facilities

     Many refiners also operate gasoline blending facilities
and/or import gasoline and thus would be considered to have
multiple modes of operation, i.e., gasoline producing, blending
and importing are each considered to be a mode of operation.
In the July 9, 1991 proposal, EPA proposed two options for
dealing with multiple modes of operation.  Under the first
option, each mode would have its own baseline (be it a unique
baseline or the statutory baseline)  and would comply with that
baseline.  EPA stated that, under this option, post-1994
shifting of gasoline and blendstocks between the modes could
occur.  Dumping could thus result if each mode were required to
comply with a separate baseline.  Under the second option, the
baselines of each mode would be combined to form a single
baseline for compliance purposes.  This option could reduce
dumping since all of a refiner's gasoline would contribute to
its baseline.  However it could also provide a refiner
significant flexibility and possibly a competitive advantage
over those with more limited operations.
     In the April 16, 1992 proposal, EPA proposed having
individual baselines for each mode to minimize anti-competitive
effects.   Integrated companies which have more than one mode of
operation would have more flexibility than smaller refineries
if allowed to have a single baseline for their entire
operation.  Further review of this issue suggested that
separate baselines for each mode might instead reduce dumping
by not allowing significant quantities of "dirtier" than
average gasoline to be produced or imported, a change from the

                               425

-------
view expressed in the July 9, 1991 proposal.  Accounting and
compliance determinations would also be less complicated if
each mode of operation were subject to its own baseline.

     In the February 26,  1993 proposal, EPA requested comment on
allowing closely integrated blending operations to be included
in the baseline determination of a refinery.  For example,  a
blending facility which receives most of its gasoline and
blendstocks from an associated refinery, and very little from
outside of the company, would be considered closely integrated.
In such a situation, the concern about anticompetitive effects
is greatly reduced.  EPA also stated that it does not expect
"closely integrated" to apply broadly to refiners who happen
also to have blending operations.  In this proposal, EPA did
not propose a similar inclusion of operations for refiners with
import operations.

     Many commenters supported allowing a single baseline for
multiple modes of operation.  Some commenters suggested that
refiners with import operations also be allowed to develop a
single baseline covering their refining and importing
operations.  One commenter suggested basing individual
baselines on whether marketing or producing gasoline occurred
at a facility.  Requiring 60-75 percent of a blending
facility's blendstocks to have come from a single refinery was
suggested for defining a closely integrated refinery-terminal
relationship.  If closely integrated, one commenter suggested
that the blending facility baseline be determined using Method
1-type data.

     For the reasons described above, EPA is not, in general,
allowing a single baseline for multiple modes of operation.
However, EPA is allowing refiners with blending facilities
which are closely integrated to include a blending facility's
production  (properties and volumes) in a single baseline.  The
criteria for being "closely integrated" is that at least 75
percent of the blendstock received at the blending facility in
1990 came from the associated refinery.  EPA believes that the
"closely integrated" criteria of 75 percent is reasonable
because oxygenates and butane, among others, are blended into
gasoline after the refinery.  Based on the usual amounts (in
volume percent) of oxygenates and butanes frequently added
outside of the refinery,  EPA estimated the "75 percent"
criteria.  For instance,  ethanol blended into gasoline in 1990
was likely blended at the 10 volume percent level.  Thus only
90 percent of the gasoline was refinery-produced blendstock
(refinery meaning the associated refinery).   While EPA could
have required 90 percent of a blending facility's blendstock to
have come from the associated refinery, in the case of ETBE, up
to 17 percent ETBE by volume could be added to gasoline.  Thus
only 83 percent of the blending facility's gasoline would have

                               426

-------
been (or will be)  refinery-produced blendstock.  If butane is
added in the winter, even less of the blending facility's
gasoline would have been refinery-produced.  Using the rule of
thumb of 2 volume percent butane per psi increase,
approximately 6 volume percent butane would be required to
increase the RVP of summer gasoline from 9.0 psi to 12.0 psi, a
not uncommon winter fuel RVP.  Thus, as little as 77 percent of
a blending facility's blendstock could be refinery produced.
The 75 percent criterion insures that a significant portion of
a blending facility's blendstock comes from the associated
refinery, but allows it some leeway in obtaining blendstocks
from other sources.

     One or more blending facilities may be included in an
aggregate baseline  (to be discussed in paragraph 5.c)  if each
blending facility received at least 75 percent of its
blendstock from one or more of the refineries in the aggregated
baseline with which it is associated.  For instance, the 75
percent criteria is satisfied if the terminal received 25
percent of its 1990 blendstock from refinery A and 50 percent
from refinery B.  Alternatively, it may also have received the
entire 75 percent from either refinery A or B, refinery A and B
being part of an aggregate baseline.  EPA believes it is
reasonable to allow this since the blending facility is still
dependent on receiving a significant portion of its blendstock
from within the company.

     Although not previously proposed, some comments were
received regarding other types of closely integrated facility
relationships.  EPA is thus allowing a single individual
baseline to be determined for two or more refineries (or sets
of gasoline blendstock-producing units) which are
geographically near each other but are not within a single
refinery gate, and whose 1990 operations were significantly
interconnected.  The burden is placed on the refiner to show
that its two facilities are "significantly interconnected".  In
this case, the two facilities will have a single set of
baseline parameter values and associated emissions.

     Some commenters suggested that U.S. refiners with import
operations also be allowed to develop a single baseline
covering their refining and importing operations.  EPA rejected
this suggestion because it would be difficult for EPA to track
a fuel's production location before the fuel is or was
imported, particularly when considering 1990 production.  Also,
allowing such a situation would amount to trading between
foreign and domestic refineries, which was not mandated nor
intended by Congress.

b.   Seasonal Weighting
                               427

-------
     In the February 26,  1993 proposal,  EPA proposed that a
refinery's own production volumes of 1990 summer and winter
gasoline  (based on whether the fuel RVP met EPA 1990 summertime
volatility specifications) be used in the weighting of data on
a summer and winter basis.  This change from the previous
proposal, in which EPA specified a summer/winter split,
received a lot of support.

     As discussed in paragraph 5.a, the 1990 annual baseline
volume is the larger of the gasoline volume produced in or
shipped from the refinery in 1990.  Thus, a refinery's own
baseline volumes of summer and winter gasoline  (either on a
produced or shipped basis) shall be used for weighting the
summer and winter anti-dumping emissions and sulfur, olefins
and T90 values.  Summer volume consists of the volume of
gasoline produced to meet the 1990 Federal volatility
requirements.  All other gasoline volume, including transition
volumes, is considered winter volume (i.e., fuel produced or
shipped during transitions from winter to summer is considered
winter data because it does meet the summer volatility
requirements).   Because fuel production can lead fuel
consumption by up to 2 months, use of refinery-specific volumes
and time periods in the baseline determination will result in
more representative baselines.

c .    Grade Weighting

     To minimize data-skewing due to inconsistent or special
sampling situations (for instance, a trial gasoline which was
heavily sampled compared to normal sampling),  and to thus get a
more representative baseline, average fuel parameter values
shall be determined first for each grade of gasoline produced,
and the resulting values weighted by the fraction of each grade
sold in the period over which the value is determined.  "Grade"
shall mean each traditional grade of gasoline produced in the
refinery in 1990, e.g., regular, midgrade, premium, not each
different integer octane number as proposed in February 1993.
EPA changed this definition based on comments received stating
that requiring each octane number was onerous and impractical.

d.    Equations

     The equations have been modified slightly from the February
1993 proposal to require that specific gravity be included in
the determination of baseline sulfur and oxygen contents.
Because both of these fuel parameters are determined on a
weight basis, and because gasoline and blendstocks vary,
sometimes significantly,  in weight-to-volume ratio, correct
accounting of such terms must include a weight-to-volume
conversion.
                               428

-------
     The equations have also been modified to reflect the
separate determination of summer and winter baseline fuel
parameter values.

e.   Oxygen in the baseline

     Congress mandated that reformulated gasoline contain
oxygen, but did not mandate a post-1994 conventional gasoline
oxygen content.127  In 1990,  some refiners produced or utilized
significant amounts of oxygenated gasoline while others used
almost no oxygen in their gasoline.  The Act requires that,
relative to 1990, increases in NOx emissions due to oxygenate
use be offset by reductions (on a mass basis) in VOC, CO and
toxics emissions.

     The issue of how to treat oxygen content in the baseline
and for compliance purposes was not addressed in the negotiated
agreement nor in the July 9, 1991 proposal.  However, the issue
was discussed in the April 16, 1992 proposal, and two viable
options for dealing with a refiner's baseline oxygen were
identified.  Under the first option, only a positive difference
between the post-1994 annual average oxygen level and the
baseline oxygen level would be included in the compliance
calculation.  If less oxygen were used in post-1994 gasoline
than in 1990, the individual baseline value would be used in
the compliance calculation,  and there would be no penalty for
using less oxygen in post-1994 conventional gasoline than in
1990.  If more oxygen were used in post-1994 conventional
gasoline than in 1990, credit would be given for the effect of
the excess oxygen on emissions.  The second option excluded
oxygen in both the baseline and compliance determinations.
Thus there would be no penalty for using less oxygen than in
1990 and no credit for using more oxygen (in conventional
gasoline) than in 1990.  If no credit is given for oxygen usage
above base levels, there may be less processing flexibility
relative to conventional gasoline production.  In that
proposal, EPA noted that Congress purposely did not include
oxygen content in the statutory definition of summertime
baseline fuel.  EPA likewise did not include oxygen content in
its determination of the proposed statutory wintertime baseline
gasoline.

     Several commenters suggested that the baseline be
determined on a non-oxygenate basis so as not to penalize those
who "reformulated", i.e., produced cleaner gasoline, early.
Others supported including only the positive difference between
     127   Certain conventional gasoline areas which are CO
nonattainment areas are required to contain oxygen during their
oxygenated fuel season.

                               429

-------
1990 and post-1994 oxygenate use.  Others suggested variations-
-excluding it in the baseline but including it in compliance;
excluding it from the baseline and normalizing the values of
the other parameters to account for the dilution effect of
oxygenate; and including it, as is, in both the baseline and
compliance calculations.  Relative to other calculations, some
commenters suggested that oxygenate volume be excluded from
conventional gasoline volumes.  Others argued that including
oxygenate used in conventional gasoline designated for areas
for CO reduction purposes would give credit in one program for
another mandated program, and that aromatics markets
distortions and increased toxics emissions and dumping could
result.

     The anti-dumping provisions of §211(k)(8)  require a
comparison of 1990 and post-1994 emissions,  based on 1990
gasoline quality and post-1994 gasoline quality, respectively.
The quality of any gasoline includes its oxygen content, thus
including oxygen content in the baseline and compliance
calculations would be the most appropriate consideration of
oxygen content.  EPA is therefore requiring baseline parameter
values to be determined on an oxygenated basis.  However,
§211(k)(8)(C) of the Act also requires that increases in NOx
emissions, due to conventional gasoline oxygenate use, be
offset by reductions in the other three pollutants.  As stated
earlier,  significant VOC and CO reductions will occur even
without the reformulated gasoline rulemaking.  To ensure that
an increase in NOx emissions is not associated with the use of
oxygen,  EPA is allowing refiners to choose to use either an
oxygenated or nonoxygenated baseline when determining NOx
emissions.  Compliance would be measured on the same basis.
Under this provision, a refiner could choose to switch from a
nonoxygenated to an oxygenated baseline, beginning with the
next averaging period.  While EPA is allowing refiners to
choose whether to use a nonoxygenated or an oxygenated baseline
when determining NOx emissions, the initial decision to use an
oxygenated baseline, or the switch from a nonoxygenated to an
oxygenated baseline is permanent.  EPA is allowing only the
onetime change to minimize unnecessary administrative
complications due to every refiner potentially changing its
baseline NOx value annually.  EPA expects a refiner to operate
its refinery to its advantage, and thus it is not likely to
make such decisions  (of whether to use a nonoxygenated or an
oxygenated baseline for NOx purposes)  lightly.   Additionally,
Congress intended that the anti-dumping program compare a
refiner's 1990 emissions with its post-1994 emissions, based on
its fuels' actual average composition, i.e,  its actual
oxygenated quality, and a change to this baseline for NOx
purposes should thus be permanent.
                               430

-------
     In determining the nonoxygenated parameter values from the
oxygenated values, only the physical dilution and distillation
effects of the oxygenate shall be considered.  For instance, if
a refiner's actual (oxygenated)  baseline aromatics were 30
volume percent and actual oxygenate use was 5 volume percent,
the nonoxygenated baseline aromatics value would be 31.6 volume
percent, or 30/(100%-5%) .   Adjustments to refinery operations
that would have been different had oxygenates not been used
(i.e., octane) shall not be included because many potential
adjustments are possible.   In the above example, while it is
likely that reformer severity may have been higher had
oxygenates not been used (thus resulting in perhaps even a
higher aromatics baseline value)  such operational effects due
to oxygenate use shall not be considered because they cannot be
known with certainty.

     A few commenters suggested that oxygenate volume be
excluded from conventional gasoline baseline volumes.  EPA
disagrees--Congress specified that certain NOx emissions
increases be offset,  but did not specify how to deal with
baseline volumes, leaving it to EPA's discretion.
Additionally, the reason for allowing NOx emissions to be
evaluated on a nonoxygenate basis in the first place is so as
not to penalize refiners whose emissions increase due to
oxygenate use.  It is possible that restricting baseline
volumes by excluding oxygenate volumes could penalize some
refiners.  Thus,  it would be inappropriate for EPA to restrict
the applicability of the individual baseline to the
nonoxygenated gasoline volume.  The total gasoline volume
(including the volume of actual 1990 oxygenate use and the
volume of those oxygenates assumed or shown to have been
blended as discussed in paragraph 2.a) will thus be used to
determine the 1990 baseline volume, as well as for the volumes
to determine applicability of the individual baseline.

f.   Work-in-Progress

     Work-in-progress (WIP)  refers to one or more major capital
changes or  commitment(s)  made by a refiner prior to or in
1990, the baseline year for anti-dumping purposes.  In those
cases, the gasoline produced in 1990 might significantly differ
from gasoline produced after the WIP was finished.  However,
since the anti-dumping baseline is based on 1990 gasoline, a
refiner's compliance for post-1994 conventional gasoline would
be measured by comparing emissions of conventional gasoline
produced after the WIP with the emissions of 1990 gasoline
produced prior to the WIP.   For some of these refiners,
compliance with the anti-dumping requirements (to not exceed
1990 emissions levels) could only be accomplished by not
operating, or not fully operating, the new or upgraded unit.
The inability to make use of the new or upgraded unit could in

                              431

-------
turn threaten the financial viability of some companies.  EPA
is allowing WIP baseline adjustments, under certain
circumstances, as discussed below.  A WIP adjustment allows a
refiner to modify its baseline volumes and fuel parameter
values such that they reflect the quality of its gasoline after
completion of the WIP.  In order to obtain a WIP modification,
a refiner must petition EPA and EPA must approve the petition.

(1)   Background

     In the July 9,  1991 proposal, EPA asked for comments as to
whether baseline adjustments should be allowed for major
construction, or associated financial commitment, which
occurred in 1990 and which would change a refinery's future
gasoline quality.  At the same time, EPA expressed concern
about the legal basis for providing the option of having an
"adjusted" baseline when the Act clearly states that a
refiner's 1990 baseline is the basis for compliance with the
anti-dumping requirements.

     WIP was also addressed in the regulatory negotiations, and
the negotiated agreement provided general guidelines on which
to base the decision for allowing a WIP baseline adjustment.
The guidelines were meant to restrict petitions for WIP
adjustments to those extreme situations where it would be
appropriate to make an adjustment.  The guidelines in the
agreement were stated as follows:

     "A refiner's baseline will be modified for refinery
     work-in-progress at year-end 1990 if it can be
     demonstrated that

     1.  such a modification was significant, i.e., if at
     least a five percent difference between baseline
     emissions calculated with and without the work-in-
     progress resulted,  and

     2.  such work was associated with other regulatory
     requirements,  and

     3.  failure to grant credit would result in an
     extraordinary regulatory burden, i.e.,  a substantial
     portion of the refinery's capital would be at risk if
     the adjustment were not granted, and

     4.  such credit would not result in significant
     environmental harm (e.g., the credit will not cause
     the refiner to be five percent above the Clean Air Act
     antidumping baseline, etc.)."
                               432

-------
     EPA's April 16,  1992 proposal further discussed baseline
adjustments for WIP.   The guidelines presented in the
negotiated agreement were included in this proposal as
requirements,  all of which had to be met in order to obtain an
adjustment.  In this proposal, adjustments were proposed to be
allowed only for significant differences in exhaust benzene
emissions.  EPA also stated that authority for a WIP baseline
adjustment could be found in Alabama Power Company v. Costie,
636 F.2d 323.357 (B.C. Cir 1979) which explained that agencies
possess "equitable discretion...to afford case-by-case
treatment--taking into account circumstances peculiar to
individual parties in the application of a general rule to
particular cases, or even in appropriate cases to grant
dispensation from the rule's operation."  Thus, in allowing WIP
adjustments, EPA would be exercising its "equitable" discretion
to afford limited relief to refiners which had work-in-progress
and which would be extraordinarily burdened by the anti-dumping
requirement that they not exceed their 1990 emissions.  EPA
proposed that the "substantial"  capital-at-risk requirement
would be met if the capital involved with the WIP was greater
than or equal to 10 percent of the refinery's depreciated plant
and equipment value.

     In the February 26, 1993 proposal,  EPA expanded upon and
clarified certain aspects of the WIP criteria.  First, EPA
proposed that the WIP adjustment also apply to the fuel
components which are capped when the simple model is used, and
to exhaust toxics and NOx emissions in the years when the
complex model is required.  The  latter emissions requirement is
consistent with the negotiated agreement since the first
guideline in the agreement states "...at least a five percent
difference between baseline emissions...".  While allowing a
WIP adjustment to apply to the fuel parameters controlled under
the simple model (sulfur, olefin and T90) was not specified in
the agreement, it is appropriate to allow such adjustments
because those fuel parameters are standards with which a
refiner must comply.   Specifically, EPA proposed that if one or
more of the following situations occurred, and all other
criteria were also met, a WIP adjustment would be allowed:
                               433

-------
     1)  at least a 25.0 percent difference exists between
     sulfur,  olefin and/or T90 values calculated with and
     without  the WIP adjustment;

     2)  at least a 5.0 percent difference exists between
     the refiner's baseline exhaust benzene emissions
     calculated with and without the WIP adjustment;

     3)  at least a 5.0 percent difference exists between
     the refiner's baseline exhaust toxics emissions
     calculated with and without the WIP adjustment;

     4)  at least a 5.0 percent difference exists between
     the refiner's baseline NOx emissions calculated  with
     and without the WIP adjustment.

     Also in  the February 26,  1993 proposal,  EPA proposed and
requested comments on allowing either the "10 percent" criteria
from the April 16, 1992 proposal or a $10 million minimum cost
of the WIP to satisfy the capital-at-risk criteria.  Comments
were received which stated that the 10 percent requirement
could be biased against recently built or recently upgraded
refineries because the depreciated value would be higher for
these facilities than for older refineries of the same size.
The commenter also claimed that newer facilities tend to
benefit the environment more than similar, older facilities,
and thus could be unfairly penalized, compared to older
facilities, by the proposal.

     EPA also proposed that WIP include projects actually under
construction in 1990 and projects for which contracts were
signed in 1990 such that construction would be completed in
time to comply with the regulatory requirement motivating the
WIP (per the  proposed requirement that stated the WIP must be
associated with other regulatory requirements).

     In the February 26, 1993  proposed regulations, EPA expanded
on the last of the criteria outlined in the agreement, that a
WIP adjustment not result in significant environmental harm.
The agreement suggested that the adjustment not cause the
refiner's baseline to be more  than five percent above the
                               434

-------
"Clean Air Act anti-dumping baseline".  In the April 16, 1992128
proposal regulations, EPA addressed this criteria as follows:

     "...Such a credit would not cause a refiner's baseline
     exhaust benzene emissions... to exceed the exhaust
     benzene emissions due to the baseline gasoline...by
     more than five  (5.00) percent.

     ...A work-in-progress-adjusted baseline exhaust
     benzene emission value which is greater than 105
     percent of the exhaust benzene emissions due to the a
     baseline gasoline... shall be allowed if the pre-work-
     in-progress baseline exhaust benzene emissions value
     was greater than the work-in-progress-adjusted value."

     As stated, EPA's 1993 proposal would expand this criteria
by creating parallel requirements to cover the exhaust toxics
and NOx emissions and simple model fuel parameters.  The
language included in the February 26, 1993 also proposed
regulations  (shown below129 in paragraph (A))  to cap exhaust
emissions and sulfur, olefin and T90 values.  Additionally, the
language shown in paragraph  (A)  capped the WIP adjustment of
refiners whose unadjusted baseline (parameters or emissions, as
applicable) was less than the statutory baseline at 105 percent
of the statutory baseline. The language shown in paragraph  (B)
and (C)  allowed refiners whose unadjusted baseline was greater
than the statutory baseline to maintain their unadjusted
baseline value (and not have to reduce it to 105 percent of the
statutory baseline).

     "...(A) Such a credit would not cause a refiner's
     baseline exhaust benzene emissions,  exhaust toxics
     emissions or NOx emissions, ... to exceed the respective
     emissions due to the baseline gasoline...by more than
     five  (5.00)  percent nor allow its sulfur,  olefin or
     T90 values to exceed the corresponding values...by
     more than five  (5.00) percent.
     128 From §80.91(c) (4) (iv)  of the March 31,  1992 proposed
regulations.  The April 16, 1992 proposal published in the
Federal Register contained only the preamble language, it did
not include the associated regulations.  However, both the
preamble and regulations were completed on March 31, 1992; the
April 16, 1992 published preamble differed from the March 31,
1992 preamble only in minor ways to allow it to be printed in
the Federal Register.  Thus the regulations associated with the
April 16, 1992 published proposal are actually dated March 31,
1992.

     129 From §80.91(d) (4) (iv) .

                               435

-------
     (B)  Work-in-progress-adjusted baseline exhaust
     benzene,  exhaust toxics,  or NOx emission
     values,...which are greater than 105 percent of the
     respective emissions due to the baseline
     gasoline... shall be allowed if the pre-work-in-
     progress baseline values were greater than the work-
     in-progress-adjusted values.

     (C)  Work-in-progress-adjusted baseline sulfur, olefin
     or T90 values which are greater than 105 percent of
     the respective baseline values... shall be allowed if
     the pre-work-in-progress baseline values were greater
     than the work-in-progress-adjusted values."


(2)   Comments

     Most commenters supported allowing WIP adjustments  (if all
other criteria were met) for significant differences between
unadjusted and WIP adjusted values of exhaust toxics emissions,
NOx emissions,  sulfur, olefin or T90.  A few commenters
suggested that different complex and simple model emissions
requirements have different comparison criteria.  Additional
comments suggested reducing the threshold values between WIP-
unadjusted and adjusted values, i.e., the difference in
emission or parameter values which would be the impetus for
desiring a WIP-adjustment.   Commenters stated that lower
differences between these two values (unadjusted and adjusted)
were also significant, and the proposed levels were rather high
and perhaps impractical.  Comments also included the
recommendation that if simple model requirements are the
impetus for the WIP, that a refiner revert back to an
unadjusted baseline when the complex model is used.

     Few comments were received regarding the requirement that
the WIP be associated with other regulatory requirements.
However,  additional clarification of the language was requested
concerning "..completed in time...to comply...", per the
February 26,  1993 definition of WIP as

     "...projects under construction in 1990 and projects
     which were contracted for and which will be completed
     in time for the refiner to comply with the regulatory
     requirement. . . "13°

     This language was included in the February proposal to
ensure that the WIP would be completed in a timely manner,
since the WIP was ostensibly being done to comply with a
        From §80.91(d)(5).

                               436

-------
regulatory requirement.  Less than timely completion would
indicate that the regulatory requirement was not a driving
factor in initiating the WIP.  Clarification was also requested
as to what types of contracts would be considered to have
committed the refiner to the WIP.  Other suggestions included
allowing WIP adjustments for work not necessarily associated
with a regulatory requirement, including WIP which would have a
beneficial effect on a refinery's overall environmental
performance.

     Several  comments reiterated a concern expressed in the
regulatory negotiation discussions, namely, that a WIP
adjustment should be an exception, i.e., few refiners should
qualify.

     As stated above, in the February 26, 1993 proposal,  EPA
proposed to allow "substantial regulatory burden" to be
fulfilled if the WIP value was at least 10 percent of the
refinery's depreciated plant and equipment value, or at least
$10 million.   Some commenters suggested that the WIP
requirements be more stringent,  ranging from opposition to any
modification of capital at risk  (presumably from the 10 percent
criteria)  to increasing the threshold value to $50 million.
Other comments suggested reducing the threshold value to $5
million (possibly a more appropriate value for small refiners)
or to 5 percent (instead of 10 percent) or eliminating any
"dollar" amount because no one should be penalized because its
investment fails to meet arbitrary time or cost criteria.
Other alternatives suggested include considering 5 percent of
the refinery replacement cost or 10 percent of the depreciated
book value.

     EPA received many comments and suggested language with
regard to the simultaneous capping of emissions and parameter
values in the February 26, 1993 proposal.  Commenters also
suggested that the five percent statutory baseline cap was too
stringent and that the requirement for showing a difference
between unadjusted and unadjusted values was inconsistent with
the comparison to the statutory baseline.  Other commenters
stated that because most new refinery projects are subject to
very stringent environmental regulations their overall
environmental performance will likely improve.

(3)  Final Rule

     EPA is requiring that a refiner meet each of three criteria
in order to qualify for a WIP baseline adjustment.  A fourth
criteria is also required to minimize environmental harm due to
WIP adjustments.
                               437

-------
     The first criteria,  the showing of significant difference
between unadjusted and adjusted values (i.e., calculated with
and without the WIP adjustment) has been changed from the
February 26, 1993 proposal, considering comments received on
this issue.  As stated, the February proposal required at least
a 25 percent difference in sulfur or olefins or T90 or at least
a 5 percent difference in exhaust benzene emissions or exhaust
toxics emissions or NOx emissions.  Upon further analysis of
fuel changes and corresponding emissions, reducing the
threshold values seemed reasonable, and neither environmentally
nor competitively detrimental.  Additionally, reducing the
thresholds did not appear to reduce the stringency of the WIP
criteria, because the proposed values may have been too
stringent, and perhaps unrealistic relative to actual refinery
situations.  In this final rule, at least the following minimum
differences must be shown between WIP unadjusted and adjusted
values:

     Minimum Differences  between Unadjusted and WIP-Adjusted
     Values
     Exhaust benzene emissions  (simple model):    2.5%
     Exhaust benzene emissions  (complex model):   2.5%
     Sulfur:
10.0%
     Olefins:
10.0%
     T90:
     10.0%
     Exhaust toxic emissions:
2 .5%
     NOx emissions:
2 .5%
     Although not previously specified,  the differences shown
above shall be calculated as follows:

     |Unadjusted Value - Adjusted Value|
     	x 10o =	%
          Unadjusted Value
     EPA realizes that if many refiners qualified for a WIP
adjustment, negative competitive effects could result.
Additionally, allowing many adjustments could indicate that EPA
has exceeded its equitable authority under Alabama Power, supra
(there is, however, no specified limit on the number of
refineries which may qualify).   EPA has set the above WIP
criteria so that adjustments are received only by those
exceptionally burdened by the combination of the reformulated
                               438

-------
gasoline/anti-dumping rules and other legislative or regulatory
requirements.

     The second criteria requires a showing that the WIP was
associated with another regulatory requirement.  Obviously,  the
wording of the agreement,  "...associated with other regulatory
requirements..." leaves a lot of room for interpretation.  EPA
is requiring that the WIP be associated with a legislative or
regulatory environmental requirement enacted or promulgated
prior to January 1, 1991.   Such requirements are not limited to
gasoline or Federal requirements.  For instance, stationary
source or other fuel quality regulations are acceptable
regulatory requirements.  However, the WIP only applies to
projects involving gasoline blendstock or distillate-producing
units under construction in 1990 or for which contracts were
signed prior to or in 1990 such that the refiner was
financially committed to permanently changing refinery
operations.  EPA believes that the contracts should have
committed the refiner to purchasing materials and construction
of the WIP.  As such, a process engineering design contract
does not commit the refiner to actually implementing the WIP
and would not be considered a WIP contract under this
provision.

     The third criteria requires a showing that the WIP cost was
equal to or greater than 10 percent of the refinery's
depreciated book value as of the WIP start-up date or had a
cost of at least $10 million.  Based on comments, "book" value
was substituted for the proposed "plant (including capitalized
engineering costs)  and equipment" value.  Commenters suggested
that "book" value was a more universally used and understood
term in the refining industry. EPA believes that both the $10
million or 10 percent criteria are fairly stringent
requirements,  and that more stringent requirements could
threaten the viability of some refiners.  Less stringent
criteria could indicate that the value of the WIP was not
actually sufficient to cause hardship should no WIP-adjustment
be allowed.  Thus,  this provision is being promulgated
essentially as proposed.

     EPA is requiring that determination and verification of the
WIP-adjusted baseline fuel parameter values be achieved using
data obtained subsequent attainment of reliable operation of
the WIP.

     Finally,  in order to minimize environmental harm due to WIP
adjustments, EPA is limiting WIP increases in baseline exhaust
benzene, exhaust toxics and NOx emissions and sulfur, olefins
and T90 values to the larger of 1) the unadjusted individual
baseline value of each emission or fuel parameter or 2) 105
percent of the corresponding anti-dumping statutory baseline


                               439

-------
value.  The "105 percent" values are explicitly stated in
§80.91.  Note that sulfur, olefins and T90 are only constrained
when compliance is determined using the simple model.  When
compliance is determined using the complex model,  the WIP-
adjusted values of these three fuel parameters are not subject
to the caps.  Because a WIP adjustment allows a refinery's
entire baseline (fuel parameters, emissions, and volume)  to be
adjusted, thus affecting all of the emissions and fuel
parameters controlled under the anti-dumping requirements, the
requirement described above should adequately minimize
environmental harm by not allowing extreme changes in emissions
or controlled fuel parameters relative to statutory baseline
gasoline.

     Finally,  while it may be true, as commenters  suggested,
that overall refinery emissions may decrease due to new
refinery projects, section 211(k)(8) does not provide EPA
authority to consider emissions other than those vehicle
emissions specified.  Hence, minimization of potential negative
environmental effects due to WIP must be relative  to the
allowed WIP adjustments.

g.   Baseline Adjustment for Extraordinary Circumstances

     In the February 26, 1993 proposal, EPA asked  for comment on
whether it should allow baseline adjustments for certain
extenuating circumstances.  If adjustments were not allowed,
ostensibly a refinery's baseline would be unrepresentative of
the gasoline it would have produced in 1990 had the
circumstance not occurred.  EPA requested comment  on what
criteria would have to be met, or proof given, to  demonstrate
that 1990 refinery operations were indeed different than if the
circumstance had not occurred.  EPA also asked for comment as
to the degree of financial hardship that might be  required to
be shown and on additional criteria which might have to be met
in order to restrict any such petition to situations falling
within the Agency's authority under Alabama Power, as discussed
in the April 16, 1992 proposal.

     EPA also requested comment as to whether two  examples of
potential extenuating circumstances should indeed be considered
extenuating.  The first example was the occurrence of downtime
which occurs every 4-5 years and which happened to occur in
1990  (nonannual turnaround).  The second example involved
disasters which shut down a gasoline blendstock producing unit
for 30 days or more.  EPA also inquired as to whether requiring
30 days downtime was appropriate.

     EPA also requested comment on whether a baseline adjustment
should be allowed for the Defense Department's phaseout of JP-4
jet fuel.  JP-4 is a naphtha/kerosene blend.  Its  replacement,

                              440

-------
JP-8, is a heavier fuel, and is produced from different
feedstocks.  The most likely alternate use of the JP-4
feedstock is in gasoline.  However, that feedstock cannot be
used in gasoline without first going through a reformer to
increase its octane to suitable gasoline levels.  Thus, the
average aromatic content of a refiner's gasoline will likely
increase relative to its 1990 values if the refiner had
produced JP-4.  Additionally, total gasoline volume produced
will likely also increase relative to 1990, and this increase
in gasoline production would be subject to the statutory
baseline, instead of the refiner's individual baseline.  In
certain instances, JP-4 was such a large portion of the
refiner's total 1990 production that the refiner may have
extreme difficulty in meeting the anti-dumping requirements and
could be forced to cease operating.

     Many commenters felt that baseline adjustments should be
allowed for the above situations as well as for others,
including:  almost meeting work-in-progress requirements;
changes in process units and crude quality; the Persian Gulf
War; strikes; neighboring unit owned by another company started
up or shutdown subsequent to 1990; loss of a significant market
(not limited to JP-4).   Others stated that because every site
is unique, baseline adjustments should be evaluated on a case-
by-case basis.  Still others suggested that EPA allow
adjustments only for small refiners, or for a certain level of
JP-4 production.  Regarding the JP-4 situation, commenters felt
that relief should be given in the anti-dumping program because
of the involvement of two governmental units in creating an
apparent conflict for certain refiners.

     Several commenters, however,  felt that no extenuating
circumstance baseline adjustment should be allowed.  Among the
reasons cited for not allowing adjustments were:  competitive
inequities; Congressional intent to account for 1990 only;
difficulty in defining extenuating circumstances; could be used
to void work-in-progress requirements.

     The 30 day downtime requirement received some support, and
not much in the way of criticism.

     After reviewing this issue,  EPA has decided to allow
baseline parameter and volume adjustments, upon petition and
approval, due to the occurrence of unforeseen, unplanned
downtime of 30 days or more of one or more gasoline blendstock
or distillate producing units due to equipment failure or
natural cause beyond the control of the refiner as well as for
nonannual turnaround of a gasoline blendstock or distillate
producing unit which occurred in 1990 and which resulted in the
minimum amount of downtime.  Fuel parameter and volume
adjustments shall be made by assuming that the downtime did not

                               441

-------
occur in 1990.  Baseline parameter and emissions values
calculated with and without the adjustment are to be included
in the baseline submission, and any adjustment must be verified
by the baseline auditor.  EPA believes that it is appropriate
to allow adjustments in these cases because the circumstances
described above would truly cause a refiner's 1990 baseline to
be different than it otherwise would have been.

     EPA is also permitting baseline adjustments for certain
refiners which produced JP-4 jet fuel in 1990.  Specifically, a
refiner which owns a single refinery which does not produce
reformulated gasoline, and whose ratio of 1990 JP-4 production
to 1990 gasoline production equals or exceeds 0.5, may, upon
petition and approval, be allowed to adjust its baseline fuel
parameter values and volume to reflect its refinery operation
had it not produced JP-4 in 1990.  EPA believes that it has
authority to allow such adjustments due to the discretion
afforded EPA by Congress.  Specifically, Congress did not
prohibit EPA from providing relief in certain instances.
Additionally, Alabama Power v. Costle131 gives EPA "case-by-case
discretion" to grant variances or even dispensation from a rule
where imposition of the requirement would result in minimal
environmental benefit but would extremely burden a regulated
party.

     As stated, JP-4 baseline adjustments will be allowed only
for those refiners which will not produce reformulated
gasoline.  While the anti-dumping requirements, in general,
apply to all conventional gasoline whether or not reformulated
gasoline is also produced,  in these specific cases no dumping
will occur due to reformulated gasoline production.  The intent
of Congress with regard to the anti-dumping program will be met
while not unduly burdening those that meet the specified
criteria.  Since both the unadjusted and adjusted baselines
must be determined, if a refiner granted such an adjustment
subsequently produces reformulated gasoline, its conventional
gasoline compliance would immediately be subject to its
original unadjusted baseline.

     JP-4 baseline adjustments are generally limited to single-
refinery refiners.  These refiners have no way to aggregate
baselines132 so as to reduce the combined burden of JP-4 phaseout
     131 Alabama Power Company v.  Costle.  636 F.2d 323.357 (B.C.
Cir 1979).

     132 As  discussed in paragraph 5.c,  a refiner with more than
one refinery may determine an aggregate baseline, i.e., a
conventional gasoline compliance baseline, which consists of
the volume-weighted emissions or fuel parameters, as

                               442

-------
and the anti-dumping requirements on their operations.  As
stated, in some cases, if no relief were granted in this area,
the viability of a refinery could be at stake.

     Finally, JP-4 must have constituted a significant portion
of a refiner's 1990 production in order for a significant
burden to exist.  As such, EPA is requiring that the ratio of
the refinery's 1990 JP-4 production to its 1990 gasoline
production must equal or exceed 0.5.  EPA believes this ratio
indicates that a significant amount of the refinery feedstock
used for JP-4 production would have to be converted in order to
produce gasoline.  In its February 1993 proposal, EPA requested
comment on what minimum portion of a refinery's 1990 production
JP-4 should have constituted for the circumstance to be
extenuating.  Several different ratio options were suggested by
commenters, as follows:  JP-4 production to total refinery
production, 20%; JP-4 production to gasoline production, 5%;
JP-4 production to gasoline production, 75%; and, JP-4
production to gasoline plus JP-4 production, 10%.  EPA's 0.5
JP-4 to gasoline ratio is in line with some of the commenters
suggestions.

     A baseline adjustment may also be obtained for a refiner
which has more than one refinery if all of its refineries
produced JP-4 in 1990 and meet the threshold criteria and none
of its refineries produce reformulated gasoline in a given
calendar year.  While such a refiner may aggregate its
refineries for conventional gasoline compliance purposes, since
there will be no dumping  (if no reformulated gasoline is
produced), EPA believes that minimal environmental harm will
result if a baseline adjustment is allowed under these
circumstances as well.  If one of the refineries produces
reformulated gasoline in a given year, the unadjusted baselines
of each of the refiner's refineries will be its compliance
baseline, or will be part any aggregate baseline in which it is
included.

     While the adjusted emissions baselines of those approved
for JP-4 adjustments are likely to be higher than their actual
1990 baselines  (primarily due to increased benzene and
aromatics) EPA expects minimal negative environmental affects.
Because the number of refineries meeting the criteria is
expected to be small and the total production of all such
refineries is also small, less gasoline is affected by any
baseline adjustments than if the criteria were less stringent.
Also, there will be no dumping by these refiners due to
reformulated gasoline production.  In this situation, EPA
believes that any negative environmental effects resulting from
applicable, of two or more refineries.

                               443

-------
the allowed adjustments are justifiably balanced by the reduced
burden on qualifying refiners.

     Although EPA is allowing baseline adjustments for the
specific circumstances described above, it in no way means this
to be a precedent to allow adjustments for actual or so-called
extenuating circumstances now or in the future.  The language
of the Act does not allow EPA to broadly permit baseline
adjustments, although under certain, limited circumstances,
particularly those which caused a baseline to be different than
it otherwise might have been in 1990, EPA is not prohibited
from allowing adjustments.  In any case, Congress' intended
representative baselines be utilized in the anti-dumping
program.  However, a baseline is neither unrepresentative of
1990, nor incalculable, because of post-1990 changes in crude
availability, fuel specifications, fuel markets, etc.  Congress
certainly knew that these types of changes could affect
baseline determinations, yet in creating the anti-dumping
requirements it did not require EPA to consider such factors in
determining baselines.  In fact, no direction was given to
account for two mandated fuel changes, Phase II volatility
control and lead phaseout.  Also, unlike work-in-progress  (a
specific extenuating circumstance occurring in 1990),
provisions for dealing with such situations, while mentioned
during the regulatory negotiation discussions, were not part of
the negotiated agreement.

     It is likely that circumstances for which baseline
adjustments are not allowed may cause a refiner's 1990 and
other pre-1995 conventional gasoline to be unrepresentative of
post-1995 conventional gasoline production and that such
circumstances may negatively affect some refiners.  However,
every refiner will be subject to future changes in markets,
fuel quality requirements, etc., all of which will affect the
refiner's gasoline quality and ability to comply with its anti-
dumping baseline.  Thus, except in extreme cases, baseline
adjustments due to post-1990 changes which affect refiners
would not be practical  (due to the myriad circumstances which
may exist) nor necessarily fair, and are definitely not
supported by the language of the Act nor the intent of
Congress.  EPA is appropriately not providing for such
adjustments.

h.   Inability to Meet the Requirements

     EPA realizes that every refinery is different with respect
to configuration, geography, crude availability, etc.  EPA also
realizes that many unique situations occurred in 1990 that are
probably not covered by the anti-dumping and baseline
development provisions being promulgated today.  While EPA's
goal with regard to baseline development is to develop

                               444

-------
individual baselines which adequately represent a refiner's or
importer's 1990 fuel, it does not want to penalize a refiner's
or importer's baseline development for trivial reasons.  Thus,
if a refiner or importer is unable to comply with one or more
of the specified requirements in its baseline development, it
may, upon petition and approval, accommodate the lack of
compliance in a reasonable, logical, technically sound manner,
considering the appropriateness of the alternative.  This
provision was not included in the proposals, but is included in
the final rule because sufficient comment was received
indicative of the innumerable details and specific
circumstances that exist in the baseline development.

5 .    Baseline Volume and Emissions Determination

a.    Individual Baseline Volumes for Refiners,  Blenders and
     Importers

     The individual baseline volume of a refiner which utilizes
Methods 1, 2 and or 3-type data to determine its baseline fuel
parameters shall be the larger of the total volume of gasoline
produced in or shipped from the refinery in 1990, excluding
volumes exported.  This provision is added because the volumes
of 1990 shipments and production could differ,  due to shipping
from inventory or to not shipping some production at the end of
the year.  Thus for each refiner, either produced or shipped
volumes could be larger.  While 1990 gasoline shipments
actually contributed to emissions, data is available (by
Methods 1, 2 or 3)  on 1990 gasoline production.  The difference
between the shipped and produced gasoline is expected to be
negligible with respect to baseline determination.  Volumes of
oxygenates blended into gasoline at the refinery and oxygenate
assumed or shown to have been blended into gasoline downstream
of the refinery, shall be included.  The baseline volume shall
be determined after all adjustments, such as for work-in-
progress or extenuating circumstances, have been performed.

     For the same reasons discussed above,  the individual
baseline volume of a blender utilizing only Method 1-type data
or having the anti-dumping statutory baseline as its individual
baseline shall be also the larger of the volume of 1990
gasoline produced in or shipped from the refinery  (blending
facility).  The individual baseline volume of an importer
utilizing only Method 1 or having the anti-dumping statutory
baseline as its individual baseline shall be the total volume
of gasoline imported into the U.S. in 1990.

b.    Baseline Emissions Calculation

     Every refiner must develop a set of individual baseline
data which include fuel parameter values, volume, emissions and

                               445

-------
blendstock-to-gasoline ratios.  As discussed in Section IV,
there are separate complex models from which to determine
summer and winter emissions.  As such, average baseline fuel
parameters must be determined separately for summer and winter.
Conventional gasoline baseline emissions (and sulfur, olefins
and T90 values) will first be determined separately, on a
summer and winter basis, using summer and winter fuel parameter
values.  As discussed in Section IV, when using the winter
complex model, RVP is held at 8.7 psi; thus, the actual average
winter RVP (for baseline or compliance purposes) is not a
factor in determining winter emissions.  The summer and winter
emissions (and sulfur, olefins and T90 values)  will then be
weighted by the respective summer and winter baseline volumes
to determine annual average baseline emissions  (and sulfur,
olefins and T90 values).  Compliance is determined in a similar
manner.

     As also discussed in Section IV, there are two complex
models--one for use prior to 2000 and one for use in 2000 and
beyond.  As such, every refinery will have two sets of baseline
total exhaust toxics and NOx emissions--one set applicable
prior to 2000, and one in 2000 and beyond.   Note that baseline
fuel parameter values and baseline volume do not change, only
the emissions determined from the baseline fuel parameters.  In
the case of NOx, it is likely that every refinery will actually
have four potential baseline NOx emissions values, depending on
whether the nonoxygenated or oxygenated set of baseline fuel
parameters is used to evaluate NOx emissions (see discussion in
paragraph 4.e).

     Regarding baseline emissions,  many commenters were
concerned about the effect of future revisions to the complex
model on 1990 baseline emissions and future compliance,
particularly should additional fuel parameters be added to the
model.  Many thought that the current  (final)  complex model
should be grandfathered for anti-dumping purposes, i.e., that
the baseline set per the final rule should always be the basis
for anti-dumping compliance, regardless of future complex model
revisions.  Grandfathering the complex model,  as suggested
could prevent reformulation of conventional gasoline.  Others
suggested EPA specify a time frame for establishing baseline
values of new parameters added to the model.  In the event of
revisions to the complex model,  EPA will promulgate additional
regulations which will consider the impact on conventional
gasoline, including consideration of lead time, cost and other
factors.
c.   Conventional Gasoline Compliance Baselines
                               446

-------
     The Clean Air Act refers to gasoline sold by a refiner,
blender or importer, but does not specify an averaging unit for
baseline determination.  EPA considered three possible options
for baseline determination--refinery basis, refiner basis, or
some combination of the two.   During the regulatory
negotiation, it was agreed that EPA would propose allowing a
refiner to elect to establish an individual refinery baseline.
In the April 1992 proposal, EPA proposed that a refiner could
choose to have an individual baseline for each of its
refineries or to have a single refiner baseline (and no
individual refinery compliance baselines),  but not a
combination of the two.

     Based on earlier comments, EPA realized that a refiner may
desire to have an individual baseline for each of its
refineries, if for example, one refinery has low baseline
emissions and the other has high baseline emissions, and the
low emission refinery will soon be producing primarily
reformulated gasoline.  If the operations of the high emission
refinery do not change, the refiner would be out of compliance
with the anti-dumping requirements because the contribution of
the low emission refinery fuel in the anti-dumping calculation
would be less.  Thus, the refiner would be penalized for
producing reformulated gasoline while maintaining conventional
gasoline operations that are very similar to its 1990
operations.  A similar situation could exist for refiners with
both California and non-California refineries.  If a refiner's
California refinery was a low emission refinery which did not
produce conventional gasoline in the future (because it was
producing California Phase II gasoline133) ,  the rest of the
refiner's gasoline would be subject to a baseline which
included the California refinery's low baseline if all
refineries had to be aggregated.

     EPA also realized, however, that allowing refiners to
simply pick a combination of averaged and individual refinery
baselines could give multi-refinery refiners an opportunity to
game the system and potentially grant them a significant
advantage over single-refinery refiners.  Thus in the proposal,
EPA stated that a refiner could not have multiple refiner
baselines, i.e., either all of its refineries were included in
its refiner baseline, or they all had their own baseline.

     Most of the comments received on this issue claimed that
EPA had not interpreted the Agreement-in-Principle's provision
correctly.  The agreement, according to the commenters, allowed
     133 California Phase II gasoline must be sold yearround in
California beginning January 1, 1996.

                               447

-------
refiners to decide how to aggregate134 their refineries'
baselines.  Some suggested that if aggregations are only
allowed as proposed, compliance with the simple model, complex
model and/or anti-dumping requirements would be difficult.  It
was also suggested that refiners should be able to re-aggregate
their refineries' baselines annually, or at least once.

     A few commenters agreed with EPA's proposal, although some
went further and suggested that all refineries should be
required to comply with their individual baselines, so as to
minimize advantages for multi-refinery companies over single
refinery companies.  One commenter stated that refinery-
specific baselines could actually increase dumping of high
emission feedstreams.

     Upon further consideration of this issue, EPA is allowing
refiners to choose to have one or more individual refinery
conventional gasoline compliance baselines and one or more
"refiner" baselines  (i.e., more than one grouping of two or
more refineries to form a compliance baseline).   All baseline
determination and compliance requirements applicable to
refiners also apply to refineries with individual baselines and
to aggregate baselines.  NOx compliance of all refineries in
the aggregate must be on the same basis, oxygenated or
nonoxygenated.  Moreover, the decision to group or not group
refineries is a onetime decision, i.e., refineries cannot be
re-aggregated annually.  Because of this, and because a
refiner's total emissions will be conserved, the possibility of
gaming will be reduced.  Thus EPA believes that allowing more
than one individual or aggregate baseline is equitable as well
environmentally sound.

     Because baseline aggregation, as well as determination of
the 1990 equivalent conventional gasoline volume (per §80.101),
depends on a refiner's 1990 baseline volume, refiners which get
the statutory baseline as their individual baseline (i.e., were
in operation for less than 6 months in 1990) do not have
baseline volumes.  Hence, the baselines of such refineries
cannot be aggregated with other refiners for conventional
gasoline compliance purposes.

     An individual baseline  (including both parameter and
emission values) must still be calculated for each refinery.
Individual baselines are required for several reasons.  First,
     134 When two or more refineries are grouped for the purpose
of having a single conventional gasoline compliance baseline,
the refineries shall be considered "aggregated", and the
resulting
baseline shall be an "aggregate" baseline.

                               448

-------
aggregate baselines are determined by volume-weighting the
individual baseline emission values (and sulfur, T90 and olefin
values under the simple model) of the included refineries.
Second, reformulated gasoline compliance under either the
simple model (for sulfur, olefin and T90) or early use of the
complex model is on a refinery basis.   Finally, individual
baselines must be known in the event that ownership of a
refinery changes.

     As proposed, EPA is also requiring that a refiner establish
individual refinery baselines under certain circumstances.
Specifically, as discussed in the July 9, 1991 and April 16,
1992 proposals, "localized dumping" could occur in certain
geographical areas.  If significant increases in toxic
emissions are occurring in a conventional gasoline area
surrounding a reformulated gasoline opt-in area, and if the
area is served by an isolated distribution system, EPA will
require refineries in the affected area to have individual
baselines for compliance purposes135.   If the refinery was part
of an aggregate baseline, the other refineries constituting the
aggregate remain in the aggregate, although the aggregate
baseline must be recalculated to reflect the loss of the
baseline of the refinery which left the aggregate as required.

     Most commenters supported this proposal.  One commenter
requested that refineries which will produce no reformulated
gasoline and which supply an isolated area be exempt from the
anti-dumping requirements.  The Act requires all domestic
refiners, blenders and importers to comply with the anti-
dumping requirements, regardless of the location of gasoline
production.  One commenter requested that this proposal not be
limited to a conventional gasoline area surrounding an opt-in
area but that it apply to all conventional gasolines areas
which may experience localized toxics emissions increases.
This suggestion was rejected, in part, because the Act directs
EPA to focus on refiners, and not on areas.  Additionally, it
would be extremely difficult to determine such areas, unless
those areas are somewhat isolated.  Also, this proposal was
discussed at length in the regulatory negotiations, and the
associated language was meant to limit its scope.  EPA believes
this is reasonable as refiners are required to maintain
conventional gasoline emissions at their own baseline levels,
and because several refiners supply most areas.  Except for the
isolated areas which are provided for by this provision, few
areas are likely to see unidirectional changes in toxics
     135 EPA's authority to establish such individual baselines
is based on section 211(c) of the Act which allows the
Administrator to regulate fuels or fuel additives to protect
the public health or welfare.

                               449

-------
emissions over an averaging period.  This provision attempts to
address a narrow problem, and EPA has appropriately limited it
to achieve this goal.

d.   Baseline Recalculation

     Pursuant to this rulemaking,  each refinery will have a 1990
individual baseline, whether that baseline is a unique
baseline, determined per the regulations, or whether it is the
anti-dumping statutory baseline.  Its individual baseline is
retained forever, whether new units are added to the refinery,
or other units are shutdown (the situation of the entire
refinery shutting down is discussed below).

     In the April 16, 1992 proposal,  EPA proposed that baselines
would be required to be recalculated under certain
circumstances.  EPA did not receive significant comment on this
issue.  The circumstances which would require baselines to be
recalculated are discussed below.   Only aggregate baselines are
required to be recalculated, although a refinery which was part
of an aggregate baseline may end up having to comply with its
individual baseline under certain circumstances.

     If a refinery was in operation for at least six months in
1990 and is later sold, and if the refinery was part of and/or
becomes part of an aggregate baseline, any aggregate baseline
affected by the sale shall be recalculated to reflect the
inclusion or exclusion, as appropriate, of the baseline fuel
parameter values, emissions and associated volumes of the
refinery being exchanged.

     In a change from the proposals,  EPA is  requiring that if
all gasoline producing units in a refinery are shut down for
any length of time, and if the refinery is part of an aggregate
baseline, the aggregate baseline shall be recalculated due to
the shutdown.  Excluding the baselines of shutdown refineries
is more consistent with the other baseline recalculation
situations.  However, if a refinery is only partially shutdown,
it must still comply with the requirements of the anti-dumping
program for any conventional gasoline  (and blendstocks, if
necessary, per the requirements of §80.102)  it produces.

     Note that if a shutdown refinery is later sold to another
refiner, its old aggregate baseline (if it was one of more than
two refineries comprising the aggregate baseline) and its new
aggregate baseline, if applicable, would be  recalculated.  If
the shutdown and later sold refinery was one of two refineries
comprising an aggregate baseline,  the remaining refinery would
then have an individual baseline--it could not be re-
aggregated, either by combining with another refinery with an
                               450

-------
individual baseline or by inclusion in another aggregate
baseline.

     Any baseline changes due to the circumstances described
above, or to similar circumstances, must be verified by a
certified baseline auditor,  and submitted  (with appropriate
documentation)  to EPA within 30 days of the event causing the
baseline fuel parameter values to require recalculation.

D.    Baseline Auditor

     In the regulatory negotiations, although not specified in
the negotiated agreement, the concept of having individual
baseline determinations verified by an independent source prior
to submission to EPA was discussed.  The independent source is
also known as the baseline auditor136.   In the July 9,  1991
proposal,  EPA briefly discussed its expectations of the
auditor, namely that the auditor will verify individual
baselines using good engineering judgement.  Also briefly
discussed were possible auditor certification options.  In the
April 16,  1992 EPA briefly discussed that it would develop, in
consultation with petroleum refining experts, auditor
certification criteria as well as technical guidelines for the
auditor to follow in its verification.  On December 15, 1992,
EPA held a workshop to discuss the roles and qualifications of
baseline auditors and certain aspects of the baseline approval
process.  Based on comments received at and following the
workshop,  as well as comments to previous proposals, in the
February 26, 1993 proposal,  EPA proposed specific requirements
that a potential auditor would have to meet to be able to
verify baselines, and also presented its specific expectations
of the auditor's verification.

     As discussed below, EPA is requiring that the auditor, and
any individuals and organizations which participate in
substantive aspects of a baseline verification process, meet
certain independence and technical criteria, and that the
auditor assume responsibility for its verification of an
individual baseline.  However, ultimate responsibility for the
accuracy of its baseline rests with the refiner.  Nonetheless,
if a refiner uses an auditor which does not meet the specified
requirements, its baseline submission will not be accepted
until it is verified by an auditor meeting the requirements.
     136 EPA recognizes that the term "auditor" usually refers to
a person who evaluates financial data.  However, in the context
of the baseline development discussed in this rulemaking,
"auditor" or "baseline auditor" refers to a person or an
organization retained to technically evaluate the baseline
determination of a refiner for correctness.

                               451

-------
Note that those refiners, blenders and importers which can and
do utilize the statutory baseline are not required to have a
baseline auditor.

1.   Auditor Qualifications

a.   Personnel

     The baseline auditor can be an individual or an
organization.  However, if it is an organization which is
retained for baseline verification, the primary analyst(s) must
be identified.  The primary analyst is the individual who must
sign the statement that is included with the baseline
submission indicating that it has reviewed the baseline
determination and that the baseline was correctly determined.
An auditor may utilize contractors and subcontractors, which
may be individuals or organizations.  There was a request for
EPA to define "auditor", "contractor" and "subcontractor" as
well as delineate the qualifications and restrictions
associated with each.  The auditor is a person or organization
retained by a refiner to meet the audit requirements of these
regulations.  EPA believes the terms "contractor" and
"subcontractor" are generally understood, and EPA intends these
terms to have these meanings.  The primary analyst of the
auditor is responsible for the verification, even if parts of
the verification were performed by individuals or organizations
it, or its organization, hired to perform those parts.  EPA is
requiring that all individuals or organizations involved in
substantive aspects of a baseline verification meet the
applicable criteria as specified in §80.92 of the regulations.

     EPA is also allowing a baseline auditor to be the same
individual or organization which determined a refiner's
baseline.  Not allowing the dual roles adds considerably to the
time required for baseline determination, since the auditor
performing the verification  (potentially the second outside
organization involved in a refiner's baseline determination)
must learn about the refinery from ground zero, whereas the
person or organization which determined the baseline will have
already established a knowledge base.  The cost savings could
be considerable for smaller refiners who do not have the in-
house capability to prepare the baseline.  Several commenters
who addressed this issue agreed that the auditor should be
allowed to also be the baseline preparer, mostly from a cost
savings point-of-view.  One commenter was concerned that the
independence of the review would be lost.  While some
independence of review may be lost, this concern is minimized
by the independence requirements discussed below, and EPA
believes that the cost and time savings are relevant
considerations.
                               452

-------
b.   Independence

     One of the primary criteria required of the auditor is that
it be independent of the organization whose baseline it is
verifying.  This requirement is intended to mitigate conflicts-
of-interest.  Any person involved in substantive aspects of a
baseline verification, including organizations and individuals
associated with the auditor, its contractors and
subcontractors, must meet the independence criteria.
"Substantive aspects" would include any function that involves
analysis, review, calculations, etc. relative to the
verification.  Thus, for instance,  clerical personnel would
likely be exempt from these requirements.  Also, individuals
within the same organization may meet the independence criteria
for some refiners but not for others.  Independence must only
be shown for the refiner on whose baseline verification the
person is working.  Persons who do not meet the independence
criteria may work on developing the baseline, but not on the
verification.

     The independence requirement is designed to avoid conflict
of interest, and thereby minimize the chance that the results
of an audit are influenced either intentionally or
unintentionally by factors other than the accuracy of the
baseline determination, such as a desire to avoid injuring an
ongoing financial or other relationship.  Another concern in
such a situation is that if the auditor also develops the
baseline, then as an auditor of themselves they may be
influenced to avoid pointing out mistakes made in developing
the baseline.  EPA believes, however, that this risk is limited
in nature, and the benefits in time and cost from allowing the
auditor to also develop the baseline outweigh the risks.

     Few comments were received relative to the criteria which
were proposed in the February 26, 1993 proposal.  Thus, EPA is
promulgating the following independence requirements.  First,
no person subject to the independence criteria shall have
worked in the past three years as an employee of the refiner or
importer whose baseline is being verified.  This does not
include having been a contractor or subcontractor to the
refiner or importer.  This is a slight modification of the
proposal, where the criterion was five years, and is based on
comments received that industry turnover is sufficiently high
that three years was also appropriate.  EPA agrees that three
years is probably sufficient to establish independence.

     The second independence requirement requires that personnel
subject to the independence criteria be financially independent
of the refiner or importer.  Thus,  no person subject to the
independence criteria shall 1) have received more than one
quarter of its revenue (a change from income) from the refiner

                               453

-------
or importer during the previous year (prior to its hire by the
refiner or importer)  and 2) have a total of more than 10
percent of its net worth (a change from total investment
holdings)  with the refiner or importer.  As indicated, some of
these requirements are slight modifications of those proposed,
based on comments suggesting more appropriate wording.  Income
received from the refiner or importer to develop the baseline
being audited (verified)  is excepted.  EPA agreed with the
suggestion of one commenter, and is thus including as a
requirement, that compensation for baseline auditing services
be independent of the outcome of the audit.

c.   Technical Ability

     Because verification of a refiner's or importer's baseline
is likely to be technically challenging, the auditor must be
technically able to evaluate a baseline determination.  Its
technical knowledge must include, but is not limited to, an
understanding of:  petroleum refining processes; gasoline
blending;  the effects of process and feedstock changes on the
quality of gasoline blendstock streams; associated
computational procedures and methods of product analysis.  It
must have data gathering and analysis skills, and financial
knowledge relative to refinery operations and capital
equipment.

     Few comments were received concerning the proposed specific
technical requirements of auditors.  In order to increase the
likelihood that an auditor will possess the necessary technical
knowledge and ability, EPA is requiring that the primary
analyst(s)  have a minimum amount of refinery-associated
experience.  Specifically,  a primary analyst must have 1)
worked at least 5 years in either refinery operations or as a
consultant for the refining and gasoline-producing industry,
and 2)  at least 3 years experience working with refinery
computer models, such as those used to evaluate unit operations
or for refinery planning and economic analyses, if such models
are used in support of the baseline determination and/or
verification  (the primary analyst must have this experience
with the specific model(s)  used in the baseline verification).
EPA expects the primary analyst to know the relevant details
and assumptions of the baseline determination it verified.  EPA
may allow,  upon petition, a primary analyst to have less than
the minimum experience if its technical ability can be
otherwise satisfactorily demonstrated.

2.   Auditor Certification

     EPA is requiring that the qualifications of an auditor be
shown in a statement which is submitted to EPA.  The statement
should show how the auditor, its contractors and subcontractors

                               454

-------
meet the independence and technical criteria.  It can also
include additional information which may assist EPA in its
review of the auditor's qualifications.  If the auditor's
qualification statement does not adequately address the
independence and technical ability criteria, the auditor will
not be certified for baseline auditing, and baseline
submissions verified by that auditor will not be accepted until
such criteria are fully addressed and the auditor is shown to
meet all the requirements.

     In the February 26,  1993 proposal, EPA proposed that an
auditor's qualification statements can be submitted with a
refiner's or importer's baseline submission, without the
auditor having first been approved by EPA.  In this case, the
refiner or importer is assuming that the auditor will meet the
requirements.  However, if the auditor does not meet the
criteria, the baseline submission will not be accepted until
the baseline has been verified by a qualified auditor.  EPA
also proposed that qualification statements can also be
submitted to EPA prior to a baseline submission by a refiner or
importer or at any time by the auditor.  If no response was
received from EPA within 30 days of application or the date the
final rulemaking is published in the Federal Register
(whichever is later) ,  the auditor would be deemed EPA-
certified.

     Most commenters agreed with allowing either refiner,
importer, or auditor submission of qualifications, as discussed
above.  One commenter thought that EPA should notify auditors
of approval rather than letting them be certified by default,
and that they should be pre-certified.  EPA believes however,
that refiners could evaluate the requirements described in the
February 26, 1993 proposal (which are not much different from
those listed in this final rulemaking) and evaluate the
credentials of the auditor they want to use to verify their
baseline determination.  Since its baseline will not be
approved unless the auditor meets the qualifications, the
refiner or importer is likely to carefully choose its auditor.
Thus EPA does not believe that precertification is necessary.
In a modification of the proposal, EPA is changing the timing
of being certified by default from 30 to 45 days.  This allows
some additional time for the qualification statement to reach
EPA, considering mailing and delivery delays, etc.

     As stated, the refiner or importer is responsible for
choosing an auditor which meets the requirements.  In the
February 26, 1993 proposal, EPA had proposed that within thirty
(30) days of hiring a baseline auditor, or the date the final
rulemaking is published in the Federal Register  (whichever is
later), each refiner and importer must inform EPA of the name,
organization address and telephone number of each auditor


                               455

-------
hired.  However, upon further review, EPA believes that this
information only need be provided to EPA once, and that
notification in the baseline submission will suffice.  The
baseline submission must include all auditors hired,  whether or
not they were retained through completion of the verification
or not.

3.    Auditor Responsibilities

     The auditor is to verify the baseline determination,  not
necessarily perform a second determination.  Based on comments
received, it did not appear necessary to require a re-
development of the baseline--verification would suffice.
During the workshop, EPA had discussed the possibility of
requiring auditors to submit a plan of their verification
process to EPA prior to beginning the verification.  EPA
believes that such a requirement would increase the time
required to prepare a baseline submission, and impose
additional costs,  and is unnecessary if the auditor
(particularly the primary analyst)  is required to meet certain
criteria.  Since EPA is specifying such criteria, and
considering the time and cost aspects, auditors are not
required to submit a verification plan.

     The auditor's primary responsibility is to insure that all
baseline determination and submission requirements are
fulfilled.  This includes, but is not limited to:  insuring
that all data is correctly accounted for; that, depending on
the circumstance,  all the appropriate criteria are met; that
calculations are done correctly; that proper adjustments have
been made when necessary, such as when using Method 3-type
data, or when adjusting for work-in-progress or extenuating
circumstances, as allowed.

     EPA is also requiring the refiner to include in its
baseline submission a statement, prepared and signed by the
primary analyst, stating that to the best of its knowledge, it
has thoroughly reviewed the sampling methodology and baseline
calculations, and that they meet the requirements and
intentions of the rulemaking, and that it agrees with the final
baseline values (fuel parameters, emissions, volumes, and
blendstock-to-gasoline ratios) listed in the baseline
submission.  As stated, the refiner or importer is ultimately
responsible for the adequacy and accuracy of its baseline
determination.  However, because EPA plans to rely heavily on
the auditor's verification, if EPA discovers that an auditor
has displayed gross incompetency or intentionally committed a
significant error in the verification process, EPA may pursue
debarment of the auditor under 18 USC 1001.
                               456

-------
     In accordance with the refiner being ultimately responsible
for the accuracy of their baseline, the refiner will be the
first point of contact for questions concerning the baseline
determination.  In certain situations,  a re-verification of all
or part of the baseline submission may be required, and the
refiner may want to involve the auditor at the start of such
discussions.

E.    Baseline Submission and Approval

1.    Baseline Submission

a.    Timing

     EPA is requiring refiners and importers to submit two
copies  of their baseline determination to EPA within 6 months
of the date the final rule is published in the Federal
Register.  Two copies will allow multiple simultaneous review
of baseline submissions.  The six month timeframe was proposed
in the February 26, 1993 proposal and received some support via
comments (and minimal negative comments).   EPA believes this
timeframe is reasonable whether baseline auditors are
precertified by EPA or whether the refiner or importer has
assumed responsibility for choosing a qualified auditor without
pre-certification.  It is also reasonable considering the fact
that many refiners have already gathered sufficient data, began
estimating their baselines, and are working with potential
auditors in anticipation of the final requirements.  If a
refiner must collect data after today,  it must submit its
baseline to EPA by September 1, 1994.  Although this date is
very close to the first day of the reformulated gasoline
program, refiners will have been able to complete collection of
the required minimum data at least 30 days prior to this date
(most likely earlier) .   It will also give EPA time to review
and approve the submissions.

     EPA will consider petitions for extensions of the
deadlines,  however, refiners are encouraged to submit baseline
submission packages as soon as possible after today.

b.    Petitions

     In many situations in the baseline determination, a refiner
or importer is required to petition EPA in order to be allowed
to account for a variance from a requirement.  In other
situations, the refiner or importer is required to "show" that
it meets certain criteria.  In either of these situations,
approval will be given by the Director of the EPA's Office of
Mobile Sources, or designee.  As will be discussed below, all
petitions must be included in the baseline submission--in fact,
in most cases, baseline calculations have to be determined both

                               457

-------
with and without the requested variance, since the outcome of
the request would be unknown.  Although not previously
proposed, EPA is allowing petitions and "showings" to be
submitted prior to the baseline submission deadline although an
early decision on the request is not guaranteed.  Nonetheless,
the baseline submission must be submitted by the applicable
deadline, whether or not EPA has decided to approve or
disapprove the request.

c.   Submission Requirements

     Baseline submissions are required from all refiners,
including blenders, and importers.  Although minimum
requirements concerning the information contained in the
baseline submissions is specified in §80.93, no set format for
the baseline submissions is specified.  EPA expects to issue a
sample "baseline submission" memorandum soon which refiners and
importers may follow in setting up their baseline submissions.

     The baseline submission must include a letter signed by the
chief executive officer of the company, or designee, which
states that the data submitted is the extent of the data
available for the determination of all the required baseline
fuel parameter values, that sampling methodology and baseline
calculations meet the requirements and intentions of the
rulemaking, and that the final baseline parameter and emission
values listed represent its 1990 gasoline, to the best of its
knowledge.  If an entity is permitted to utilize the anti-
dumping statutory baseline parameters as its individual
baseline parameters, and does so, it must also include a letter
signed by the chief executive officer of the company, or
designee, indicating that insufficient data exists for a
baseline determination by the methods allowed for that entity.
Where an auditor is required to verify a baseline, the
auditor's statement of verification, and the auditor's
qualifications if not previously certified by EPA, must be also
be included in the submission.

     The baseline submission must contain information relevant
to the development of the baseline value for each of the
following fuel parameters:  benzene; aromatics; olefins;
sulfur; T50; T90; E200; E300; oxygen, including oxygenate type;
and RVP.  It must also contain the baseline emissions values
and baseline volumes.  Additionally, the blendstock-to-gasoline
ratio discussed in Section IX of the Preamble and defined in
§80.102 must be determined for each calendar year 1990-1993.
Although this requirement was not included in the proposals, it
makes sense to include it in the baseline submission, and have
the auditor verify it as for any other baseline data, because
future compliance will be based on this ratio.
                               458

-------
     Based on comments to its proposals,  EPA has determined that
a number of its proposed baseline submission requirements, such
as parameter measurement dates and extensive crude oil
information, were not pertinent to the baseline determination.
EPA thus revised its requirements; the minimum information
which must be included in the baseline submission is described
in §80.93.  Certain information on crudes and refinery unit
operations is still required because EPA may wish to evaluate
baseline submissions using a refinery flow simulation system.
EPA may require additional data to aid its review of the
baseline submission.  A refiner or importer may also include
additional information which may be useful to EPA in its
evaluation of the baseline submission.  In addition to the
numerical data requirements, EPA is requiring that a narrative
of the development of each baseline parameter value be included
in the baseline submission.  The narrative should discuss and
explain the sampling and calculation methodologies, as well as
discuss technical judgement and reasoning.  Such a narrative
will assist EPA in its review of the baseline submission.

2.    Baseline Approval

     EPA had proposed that it would publish baseline parameter
values, by refinery, for public comment while it reviewed
baseline submissions.  Based on comments and further review of
this issue, EPA believes that no substantive comments could
result from publishing such information because of the
complexity of the baseline determination.  Additionally, EPA
realizes that certain aspects of the baseline determination
must necessarily remain confidential in order to prevent
serious, negative competitive effects.

     Several issues were raised in the comments concerning
confidential treatment of information submitted by refiners.
Issues of confidential treatment for any information submitted
is regulated by EPA's Confidential Business Information  (CBI)
regulations, 40 CFR Part 2 Subpart B.  Under the Freedom of
Information Act (FOIA),  a person has a right to release of
documents unless the documents are covered by an exemption such
as the CBI exemption.  A lot of the information submitted by
refiners for baseline approvals might well be CBI, and would be
protected under CFR 40 Part 2.

     The CBI claim should be made with the baseline submission,
or it can be released without further contact with the
submitting party (40 CFR 2.203 (a)(c)).   In accordance with 40
CFR Part 2, Subpart 2, any person or organization providing
information to EPA in connection with the determination of a
baseline, including establishing a baseline or investigating
baseline determination discrepancies, may assert that some or
all of the information submitted, except the baseline emissions

                               459

-------
or parameter values which are the standard for a refiner,
refinery or importer, is entitled to confidential treatment.
Such confidential information must be clearly distinguished
from other information to the greatest extent possible,  and
clearly labeled "Confidential Business Information."
Information covered by a claim of confidentiality will be
released by EPA only to the extent allowed by procedures set
forth in 40 CFR Part 2, Subpart B.

     Because an entity's baseline will become its standard for
compliance with the anti-dumping, and possibly reformulated
gasoline,  requirements, upon Agency approval of a baseline, the
baseline exhaust benzene, exhaust toxics and NOx emissions
values and 125 percent of the baseline sulfur, olefins and T90
values will be published in the Federal Register for each
applicable gasoline producing or importing facility of a
refiner, blender or importer.  The standards themselves are not
entitled to confidential treatment under 40 CFR 2.301 (e)
(special CBI rules applicable to Clean Air Act cases).   Since
EPA must release them upon request, and because commenters did
not provide any clear or compelling reason for not publishing
them, and because there are benefits from publishing them
(citizen suit enforcement, more information to the general
public about EPA's standards, better deterrence to
noncompliance),  EPA should publish them.

     Several commenters suggested publishing a refiner's or
importer's anti-dumping index (ADI), a ratio of the individual
baseline emissions to the emissions due to the statutory
baseline.   However, there is little difference between this
value and the actual value if the statutory baseline emissions
are known.  Another suggestion included providing such
information only upon request.  Again, there is little
difference between "on request" and publishing such information
at one time.  One commenter stated that no where in the statute
was publication of baseline data required.  As discussed above,
no compelling reasons for not publishing a refiner's standards
were presented by commenters.  Another commenter added EPA
could use the "rule of three" which has been used by DOE;  in
this method, three numbers are averaged and the average
reported,  however, none of the individual numbers are publicly
available.  Although in many instances this method may be
useful, such as in reporting fuel production by grade,  in this
case, EPA does not believe its obligation to the public would
be fulfilled using the "rule of three" since each refiner has a
unique standard with which it must comply, and which will be
publicly available.

     EPA cannot specify a timeframe for approval of baseline
submissions because it cannot regulate itself.  EPA realizes,
however, the importance of having an approved baseline as far

                              460

-------
in advance of the first day of the reformulated gasoline
program (January 1,  1995)  as possible.  As such, it will
endeavor to review and approve all baselines submitted within
the appropriate timeframe in a timely manner.  The baseline
determination is not a "one-try" effort; EPA will work with the
refiner or importer until it is satisfied that the baseline is
an accurate reflection of the 1990 gasoline of the refiner or
importer.

     If a baseline has not been approved by EPA prior to January
1, 1995, then a refiner or importer may still introduce
reformulated or conventional gasoline into commerce, with
certain limitations.  First, the refiner or importer would not
be able to use the complex model early, under §80.41, as early
use of the complex model requires knowledge of the individual
baseline to determine the appropriate performance standard.
Once an individual baseline was approved, however, then early
use of the complex model would be allowed.  Second, both the
reformulated gasoline and anti-dumping programs have standards
based on the annual averages of either emissions or certain
fuel parameters.  For example, simple model standards for
reformulated gasoline require that the annual average for
sulfur, olefins and T90 not exceed the baseline values for
these parameters.  The anti-dumping standards require that the
annual average for these parameters not exceed 125% of the
baseline parameter values, and in addition, the average annual
emissions for exhaust benzene, exhaust toxics and NOx may not
exceed the baseline values for these emissions.  A refiner or
importer may introduce conventional gasoline into commerce in
conventional gasoline areas prior to approval of baseline of
the individual baseline, however, the qualities of this
gasoline will be used to determine compliance with these
various annual averages.  Failure to have a baseline approved
prior to January 1,  1995 could therefore make it harder to plan
for and comply with these annual anti-dumping standards.
Finally, failure to have an approved baseline may in certain
circumstances make it impossible to certify reformulated
gasoline under §80.41.  In such a case, the gasoline would be
considered conventional gasoline and could not be introduced
into commerce in any area covered by the reformulated gasoline
program.  Refiners and importers are urged to submit accurate,
clearly described baseline determinations as soon as possible.
                               461

-------
VIII.     Compliance with the Regulatory Flexibility Act

     The Regulatory Flexibility Act (RFA)  of 1980 requires
federal agencies to examine the effects of the reformulated
gasoline regulation and to identify significant adverse impacts
of federal regulations on a substantial number of small entities.
Because the RFA does not provide concrete definitions of "small
entity," "significant impact," or "substantial number," EPA has
established guidelines setting the standards to be used in
evaluating impacts on small businesses137.   For purposes of the
reformulated gasoline regulations, a small entity is any business
which is independently owned and operated and not dominant in its
field as defined by SBA regulations under section 3 of the Small
Business Act.

A.   Determination and Justification of Small Entity Impacts

     The Agency has found that the reformulated gasoline and
anti-dumping regulations may possibly have some economic impact
on a substantial number of small refiners.  However, these
regulations may not significantly affect gasoline blenders,
terminal operators, service stations and ethanol blenders under
the same EPA criteria.  Small business entities are not required
by the Clean Air Act to manufacture reformulated gasoline.
Since most small refiners are located in the mountain states or
in California,  which has its own  (more stringent) reformulated
gasoline program, the vast majority of small refiners are
unaffected by the federal reformulated gasoline requirements.
All businesses  (both large and small)  maintain the option to
produce conventional gasoline to be sold in areas not obligated
by the Act to receive reformulated gasoline or those areas which
have not chosen to opt into the program.

     All refiners will be affected by the anti-dumping
requirements, which are less stringent than those for the
reformulated gasoline portion of the program.  The anti-dumping
regulations affecting conventional gasoline are not expected to
disproportionately impact small refiners of conventional
gasoline.
     137U.S. Environmental Protection Agency,  Memorandum to
Assistant Administrators, "Compliance  with the Regulatory
Flexibility Act," EPA Office  of  Policy,  Planning,  and
Evaluation, 1984.  In addition,  U.S. Environmental Protection
Agency, Memorandum to Assistant  Administrators,  "Agency's
Revised Guidelines for  Implementing  the  Regulatory Flexibility
Act," Office of Policy,  Planning,  and  Evaluation,  1992.

                               462

-------
     In addition, all refiners have the option to use either the
simple or complex model during the first years of the
reformulated gasoline program.  The complex model provides
refiners with the ability to adjust the levels of various fuel
parameters such as RVP, oxygen, aromatics, olefins,  benzene,
sulfur, and fuel distillation parameters in meeting the emission
performance requirements of reformulated gasoline.   As a result,
refiners have greater flexibility under the complex model than
under the simple model  (which focuses primarily on volatility
control) in choosing the least-cost method of compliance.
Because refineries differ in process unit configuration, raw
material requirements, and product mix, the additional
flexibility afforded by the complex model should reduce the cost
of producing reformulated gasoline.  Such a situation is expected
to provide a more equitable situation for small refiners,
especially those with conversion equipment.

     The component of the reformulated gasoline program most
likely to unfavorably impact small entities is the fundamental
necessity that reformulated gasoline meet more stringent emission
standards and thus processing requirements.  The Agency is
unaware of any alternative options which might relieve the
regulatory burden on small entities while simultaneously
maintaining the program benefits required by the statute.
Exempting small refiners from the reformulated gasoline
regulations would result in the failure of meeting CAA
performance standards, which is illegal.  All reformulated
gasoline is required to meet the same performance and
compositional standards.  Additionally, enforcement of a
reformulated gasoline program  (with exemptions or less stringent
standards),  in-use,  would be impossible to enforce due to the
inherent nature of the fungible gasoline distribution system in
existence.

     The Agency interpretation of the Regulatory Flexibility Act
requires the consideration and promulgation of alternatives or,
in lieu of regulatory options, the justification for declining to
promulgate alternatives.  Nonetheless, any flexibility, in
regards to the performance standards, granted to small business
entities would unfairly provide a competitive advantage over the
business entities not classified as "small".  One available
option is reducing,  but not eliminating, regulatory requirements
on small entities.  Requirements could be scaled to just offset
any competitive disadvantage.  However if EPA were to promulgate
such guidelines, the small entities that would be eligible to
meet these "less stringent" standards 1) would be hindering
enforcement,  as a result of the fungible gasoline distribution
system mentioned above, and 2) would not be meeting statutory
provisions,  such as the fifteen percent reduction in VOCs and
toxics.
                               463

-------
     It is worthy of note that although EPA has received several
comments which claim that the reformulated gasoline regulations
will result in closing the small business entities affected by
this rule, convincing evidence supporting this claim has not been
submitted.  Despite the lack of persuasive documentation and the
inability to exempt small businesses from the requirements of the
reformulated gasoline program, EPA has made accommodations where
possible and appropriate.  One example of the versatility
embedded in the reformulated gasoline regulations, by EPA, is the
flexibility available to all refiners, both small and large, to
choose to have one or more individual refinery conventional
gasoline compliance baselines and one or more "refiner" baselines
(i.e. more than one grouping o f two or more refineries to form a
compliance baseline).   Another example of the flexibility of the
regulations is the ability to produce reformulated gasoline on a
per gallon or averaging basis.  Also, certain refiners who
produced JP-4 jet fuel in 1990 may be able to adjust their
baselines so as to reduce the compliance burden.

     In accordance with Section 604 of the RFA,  EPA has prepared
the following regulatory flexibility analysis which includes a
summary and assessment of the issues raised by public comments on
the Initial Regulatory Flexibility Analyses contained in the
previous proposals.


B.   Summary and Assessment of Issues and Comments

1.   Background

     As part of the Administrator's effort to ensure that the
regulations do not unfavorably affect small business entities,
most small business entities which will be affected by this
rulemaking were represented in the negotiated rulemaking
processes which contributed, in part, to the development of this
final rule.  The following organizations which represent in whole
or in part the interests of affected small businesses were formal
participants in the negotiated rulemaking process and signatories
to the agreement in principle:  National Corn Growers
Association, Renewable Fuels Association, Oxygenated-fuels
Association, Rocky Mountain Refiners Association, National
Council of Farmer Cooperatives, Society of Independent Gasoline
Marketers of America,  Petroleum Marketers of America Association,
Independent Liquid Terminals Association, American Independent
Refiners Association,  and National Petroleum Refiners
Association.

     EPA believes that the participation of these parties has
helped assure adequate consideration of the special position of
smaller entities in the marketplace.  During the negotiated
rulemaking sessions, the public hearings on the previous
proposals, and the public comment period, the small businesses

                               464

-------
which are potentially affected by the rule made their interests
known.  A summary of the comments are presented below.
Interested readers are referred to both dockets (A-91-02 and A-
92-12) of this rulemaking for transcripts of the hearings and
copies of all written comments.

2.    Oxygenate Treatment

     Many of the affected small businesses were concerned about
the treatment of oxygenates.   Some produce oxygenates and were
thus concerned that EPA not favor certain oxygenates in the way
that it ensured that there would be no NOx increase.  The
evidence about the effect of oxygenates on NOx emissions was not
complete at the time of the April 1992 supplemental proposal.
Some testing had been done which showed that while oxygen in fuel
lowers CO and VOC emissions,  at some concentrations and in some
forms it may raise NOx emissions.  When additional data became
available, however, there did not appear to be any significant
difference between the NOx emission effects of oxygen from
different oxygenates.  Furthermore, it appeared that reducing the
concentration of a number of additional fuel parameters
(aromatics, olefins, sulfur,  etc.)  could reduce NOx emissions.
Consequently, under the regulations being promulgated today, all
oxygenates are treated equally.  Refer to Section II of the
preamble for a detailed discussion.

     Though all oxygenates are treated equally in the
regulations, VOC and toxics emission performance will be
different for different oxygenates.  Refiners, blenders, and
importers must take emission performance into account when
reformulating gasoline with various oxygenates.  The market for
ethanol is still expected to grow as a result of the reformulated
gasoline program.  Consequently, ethanol market growth will not
pose an economic hardship on small or large ethanol producers
(refer to Section I of the RIA).  It is likely that terminal
operators not affiliated with major oil companies would be more
likely to do ethanol blending than terminals associated with
large refiners.  However, EPA has addressed the needs of blenders
in its reformulated gasoline enforcement schemes by enabling them
to blend oxygenates with reformulated gasoline blendstocks
formulated for downstream oxygenate blending.  As a result there
is no need for blenders to separately certify that they produce
reformulated gasoline and/or make baseline comparisons as
appropriate.  Ethanol blenders will need to change the way they
currently do business (unless they choose not to produce
reformulated gasoline) ,  but all blenders must do this if
reformulated gasoline performance standards are to be met.
3.    Definition of Domestic Capacity
                               465

-------
     Small businesses expressed concerned, through comments, with
the definition of domestic capacity to produce reformulated
gasoline.  Some suggested definitions that would require all
constituents of reformulated gasoline to be domestically produced
or available at domestic ports.  While some of the commenters
were concerned with creating a strong domestic market for their
oxygenates, many more were concerned that the definition be as
broad as possible.  If domestic capacity were broadly construed,
insufficient supply would not permit EPA to extend the start of
the program to opt-in areas.  As commenters explained it, an
unexpected extension might make a small refiner's investments in
improvements less valuable.  Small refiners have less of an
ability to absorb risk or to obtain financing for investments
than do their larger competitors.  Thus, small refiners want a
broad definition.  EPA agrees that it is in the interest of
domestic oxygenate producers to have a limited definition in
order that demand for their supplies be high.  Oxygenate
producers, as indeed all gasoline and gasoline component
suppliers, will be able to operate most efficiently if
uncertainty about opt-ins is minimized.

     EPA has not at this time provided a specific definition of
domestic capacity.  Furthermore, there is no fixed definition of
that term in the regulatory negotiation agreement or the Clean
Air Act.  According to §211(k)(6)(B),  EPA must consult with the
Secretary of Energy before determining whether an area has
insufficient domestic capacity.  If the Administrator finds that
there is insufficient domestic capacity to produce reformulated
gasoline, the start of the program may be delayed in opt-in areas
for up to three years.  Parties wanting EPA to perform an
evaluation of domestic capacity for a specific area should make a
formal request to the EPA.

4.    Attainment Area Blending of Ethanol

     Prior to publication to the 1992 SNPRM, EPA was concerned as
to whether there could be an impact on terminal operators and
gasoline distributors who currently blend unfinished gasoline
components with ethanol for attainment area use.  If these
blenders were not producing their own ethanol, the Agency
surmised that the cost of ethanol to attainment area distributors
would rise due to use of ethanol in reformulated gasoline.  There
will be no associated rise in the price they will receive for
their gasoline since they sell into unaffected areas.  At the
time of the April 1992 supplemental proposal, EPA invited comment
as to the extent to which these blenders do not produce ethanol.
Since EPA has not received any comments to date on this issue,
the reformulated gasoline program was developed and finalized
with the assumption that such terminal operators and gasoline
blenders were not detrimentally impacted by the regulation.

5.    Refiner/Refinery Averaging

                               466

-------
     Several small business commenters are refiners with only one
refinery.  These commenters were concerned with the ability of
multi-refinery refiners to average their baseline fuels across
many refineries for the anti-dumping provisions.  They believe
that such an averaging provision gives a competitive advantage to
large refiners.  As discussed in Section 	,  refiners are allowed
to determine the baseline for their refineries on a refinery by
refinery basis, a refiner basis, or some other basis.  If one
refinery's baseline is determined based on its own data,  then the
calculation of the baseline for the other refineries of that
refiner must not include that refinery.

     The smaller refiners were also concerned that there be some
variance procedure in cases where they serve a future
reformulated gasoline area and could not produce reformulated
gasoline through no fault of their own.  Since these refiners
generally have only one refinery and can often supply only one
market, they are more likely to be unable to supply reformulated
gasoline than a multi-refinery refiner with refineries proximate
to the pipeline.  Fortunately, the regulations provide a
mechanism for sale of conventional gasoline in covered areas
under certain very specific conditions.  The smaller refiners
acknowledged during the regulatory process that strict adherence
to these conditions (i.e., no fault, return of economic
advantage, continuing efforts, etc.) were necessary to avoid
abuse of the provision.

6.   Phase II Standards Impact on Small Refiners

     Small refiners were worried that compliance with the Phase
II reformulated gasoline requirements would be more costly than
compliance with the Phase I requirements because of the greater
VOC reduction required.  Small refineries which choose to produce
reformulated gasoline are likely to experience higher costs per
quantity of product produced than large refineries.  For
instance, refining costs in PADD IV, which tends to have much
smaller refineries than the other PADDs, are 6-7 cents per gallon
higher than for PADD III and V refineries and 3-4 cents higher
than for PADD I and II refineries138.  As stated in the A/0
Economics Bulletin139,  because of their size,  PADD IV refineries
get less benefit from economies of scale for capital investments.
     138 "Costs of Alternate Gasoline Reformulations,  Results of
U.S. Refining Study for Economics  Committee of the Auto/Oil Air
Quality Improvement Research  Program,"  Turner, Mason & Company,
April 1992.

     139 "Estimated Costs of Modifying Gasoline Properties,
Economics Bulletin No. 2," Auto/Oil Air Quality Improvement
Research Program, January 1992.

                               467

-------
However, this implies that the small refineries are not
technologically limited by the more stringent requirements.
Also, there are very few ozone non-attainment areas in PADD IV
and none have yet opted into the reformulated gasoline program.
Thus, PADD IV refineries are not expected to be impacted by the
Phase II standards.  In addition, significant relief exists for
all refiners, in that the Phase II requirements do not go into
effect until the year 2000.  Thus, approximately six years of
planning and lead time are available for determining the
necessary refinery changes.

     It is possible that using additional oxygenate in
reformulated gasoline may be a cost-effective method of meeting
the proposed standards.  Some small refiners expressed concern
that reformulated gasoline requirements would place them at a
disadvantage in producing reformulated gasoline if it limited
their ability to use ethanol.   However, a Sobotka analysis140 of
this situation concluded that small refineries are not
disadvantaged by the federal reformulated gasoline requirements
(at current opt-in levels) because they either: (1) do not
manufacture gasoline, or  (2) do not ship a significant portion of
their gasoline output to the affected markets, or  (3) could
readily serve conventional gasoline markets.

7.   Phase II Standards Impacts on Oxygenate Producers

     In the 1993 NPRM, EPA was concerned that Phase II Standards
would decrease the demand for oxygenates in reformulated
gasoline.  It is possible that using additional oxygenate in
reformulated gasoline may be a cost-effective method of meeting
the proposed standards.  In that case,  oxygenate demand should
increase and oxygenate producers should not be adversely affected
by the Phase II standards.  It is possible that the relative use
of various oxygenates could shift from their Phase I usages,
because, as stated above, the Phase II proposed standards will be
more stringent and will have a slightly different baseline
emissions level.  However, the use of all oxygenates is still
expected to be greater after 2000 than prior to 1995.


8.   Regulation Development Representation Deficiency

     The Agency also received comments from trade associations,
with small entities as members, that were not represented on the
regulatory negotiation committee for reformulated gasoline, such
as the American Waterways Operators (AWO) and the National Tank
     140 "Evaluation of the Use of Ethanol  and MTBE in
Reformulated Gasoline,"  Prepared by Sobotka & Co., Inc. for
U.S. EPA under Contract  No.  68-W9-0077,  September 30, 1992.

                               468

-------
Truck Carriers (NTTC).   As representatives of non-signatory
parties to the Agreement-In-Principle, these organizations
postulate that EPA did not consider the impact of the proposed
regulations on their industries.  Although EPA recognizes that
NTTC and AWO were not part of the regulatory negotiation, there
has been ample opportunity to testify at public hearings and
submit comments on the previous proposals to the Agency.  In
addition, many of their interests were raised through other
organizations who require their services.  Their issues are
similar in some respects to the pipeline and barge industries.
Demand for these industries'  services may increase,  if the need
to transport reformulated gasoline to opt-in areas is present.
EPA has analyzed comments submitted during the rulemaking
process.

     EPA also received comments stating that small independent
developers and marketers of oxygenated fuels were not represented
in the regulation development process, and should have been
included.  EPA believes that small independent developers and
marketers of oxygenated fuels were represented in the regulation
development process.  Specifically, the American Methanol
Institute (AMI),  the Renewable Fuels Association (RFA),  the
National Corn Growers,  the National Council of Farmer
Cooperatives (NCFC is made up of small blenders, producers and
marketers),  and the National Petroleum Refinery Association
(small refiners make up a portion of NPRA's membership)  were all
participants in the Reg-Neg process.  EPA believes these
associations adequately represented small developers and
marketers of oxygenated fuels.  In addition, the small
independent fuel oxygenate developers and marketers who commented
did not raise any issues related to the success or failure of
their businesses which have not already been addressed.

9.    Enforcement Alternatives

     NTTC states that the cost of the reformulated gasoline
program to the tank truck industry was not considered when the
rules were being developed for reformulated gasoline.  NTTC
further claims that the February 1993 NPRM favors large oil
companies private carriers and jobbers at the expense of common
carriers.  They believe this because the common carriers will not
receive any economic benefit from the ultimate sale of the
reformulated gasoline product.  Private carriers and jobbers
would benefit if the product not meeting the reformulated
gasoline specifications is sold, whether the uncertified sale was
intentional or not.  They are particularly concerned that the
presumptive liability provisions alone will unduly burden tank
truck carriers because they will be forced to pay penalties for
reformulated gasoline specification violations over which they
have no control.   As a result, EPA's final rule makes carriers
liable only where the carrier has caused a violation or where a
violation is found at the carrier's facility.  The scope of the
recordkeeping requirements for each type of party (e.g., refiners
vis-a-vis retailers of gasoline),  and therefore the cost to the

                               469

-------
party, reflects only that party's opportunity to create or alter
the characteristics and quality of reformulated gasoline.  Thus,
the smaller facilities, such as truck carriers and retailers,
have a minimal burden under the regulations whereas refiners and
importers have significant requirements.

     Some commentors stated that the costs of independent
sampling and testing will be disproportionately high for small
refiners, because their batch sizes are small in comparison to
batch sizes for larger refiners, and because independent labs may
not be conveniently located relative to small refineries,
requiring long distance sample shipping.  It is true that the
per-gallon costs of independent sampling and testing will be
larger for a refinery producing reformulated gasoline in small
batches in comparison to the per-gallon costs for a refiner
producing larger batches.  Nevertheless, EPA believes this cost
difference is insignificant.  For a 20,000 barrel batch, a small-
sized batch, the per-gallon cost of independent sampling and
testing would be $0.0003; for a 50,000 barrel batch, the per-
gallon cost would be $0.0001141.   EPA anticipates that samples
collected at refineries located distant from any reliable
independent laboratory will be shipped to the laboratory, but
does not believe such sample shipping is costly.  These
conclusions are based on EPA's experience in conducting gasoline
quality inspections throughout the country over at least the past
dozen years, when its inspectors have shipped several thousand
samples per year to EPA's laboratory for analysis.

10.  Disproportionate Regulatory Burden

     The American Independent Refiners Association  (AIRA),  a
national trade association comprised of small and independent
refiners, accounts for twenty to thirty percent of domestic
gasoline production.  It claims that Congress and several federal
agencies, EPA included, have previously recognized the importance
of small and independent refiners to a competitive marketplace
for petroleum products.  AIRA contends that the small,
independent sector of the refining industry is particularly
vulnerable to the burdens of the previous proposals because they
have limited processing flexibility, economies of scale, and
limited availability of capitol.  As a result, small, independent
refiners will experience disproportionate negative disadvantages
vis-a-vis major refiners because they have less throughput over
which to spread the costs precipitated by the requirements of the
proposed rule.  In regards to producing reformulated gasoline, if
capitol, production, and compliance costs to small refiners are
extremely high and burdensome, they need not produce reformulated
gasoline.
     141EPA estimates the cost to collect  and  store  a  sample will
be about $230, and the lab analysis  costs will be about $42
(based on an analysis cost of $415 and analysis of  10% of the
samples collected at a refinery), for  a total of $272.

                               470

-------
11.  Renewable Oxygenate Provisions

     Finally, AIRA criticized EPA's assessment of the Renewable
Oxygenate Program (ROP) in regards to small refiners.  They
perceived that neither of the Initial Regulatory Flexibility
Analyses included in the prior proposals nor the 1993 Draft
Regulatory Impact Analysis adequately considered small and
independent refiners.  AIRA specifically commented that the
requirements to produce sub-RVP gasoline would disproportionately
disadvantage small and independent refiners.  EPA is not
promulgating the renewable oxygenate provisions published in the
February 26, 1993 proposal,  and instead is basing the final rule
on the provisions as proposed in the April 16, 1992 proposal,  as
discussed in Section I of this document or Section II of the
preamble for reformulated gasoline program.
                               471

-------
IX.  State Opt-in Provisions

     The 1990 Clean Air Act Amendments (the Act) require that by
January 1, 1995, Phase I reformulated gasoline be made available
in the nine ozone nonattainment areas having a 1980 population in
excess of 250,000 and having the highest ozone design values.  In
the year 2000 Phase II of the reformulated gasoline program
begins, requiring even lower emitting gasoline.  In an attempt to
expand the benefits of the reformulated gasoline program,
§211(k)(6) of the Act allows areas, other than those mandated, to
choose to participate in the reformulated gasoline program.
Areas which so choose are considered to have "opted into" the
reformulated gasoline program.  Several key issues regarding the
opt-in provisions are discussed below.

A.   General Provisions

     Section 211(k)(6) of the Act specifies that upon application
of the Governor of a State, the Administrator shall allow any
area in the State classified as a Marginal, Moderate,  Serious, or
Severe Area for ozone nonattainment to opt into the reformulated
gasoline program.  The Act also states that opt-in shall become
effective on January 1, 1995 or one year after such application
is received, whichever is later, and that EPA shall publish each
application in the Federal Register upon receipt.

     Section 211(k)(6) grants EPA the authority to delay
reformulated gasoline in opt-in areas for up to three one-year
periods based on insufficient supply.  Such a delay could be
based on a petition from an outside party, or at EPA's own
initiative.  At the present time, EPA has yet to receive any such
petitions and believes that there will be more than sufficient
supply of reformulated gasoline for a January 1, 1995
implementation date,  given the current level of opt-in.  If a
delay is necessary, EPA must delay implementation for areas with
a lower ozone classification before issuing any such delay for
areas with a higher classification.

B.   Phase I Opt-in only

     In its February 26, 1993 proposal, EPA requested comments on
an option that would permit states to opt into only Phase I of
the reformulated gasoline program and maintain the Phase I
requirements beyond the year 2000 when Phase II of the
reformulated gasoline program begins.  EPA received a number of
comments and upon review has decided that opting into Phase I of
the reformulated gasoline program alone shall not be permitted.

     The Act does not compel the EPA to permit Phase I-only opt-
in.  Section 211(k)(10) defines reformulated gasoline as any
gasoline which is certified by the Administrator as complying
with subsection 211(k).  Per §211(k)(3)(B), the definition of
reformulated gasoline is set to change in the year 2000 from its

                               472

-------
Phase I definition to the more stringent Phase II requirements.
Thus, in the year 2000,  Phase I gasoline will no longer be
considered reformulated gasoline under §211(k) requirements.

     EPA received many arguments for and against allowing Phase
I-only opt-ins.  Those in favor argued that more areas would
likely opt into the program because the Phase I gasoline is
better defined, and the economic burden on the consumer would be
more predictable and not as onerous as with Phase II.
Nonetheless, a number of areas have already opted into the
reformulated gasoline program.  Supporters also contend that
§211(k) of the Act does not preclude Phase I-only opt-in, so it
should be allowed.  However, as discussed above, the definition
of reformulated gasoline in fact changes in the year 2000, thus
presenting a regulatory hurdle.  Some comments pointed out that
under the authority given in §211(c)(4)(C),  States could require
Phase I gasoline standards separately after the year 2000 as part
of a State Implementation Plan (SIP).   That section also states
that the Administrator may approve such SIP provisions only if
EPA determines that such State control or prohibition is
necessary to achieve the national primary or secondary ambient
air quality standard.

     Those opposed to Phase I-only opt-in were primarily
concerned with fuel proliferation,  or the possibility that both
Phase I and Phase II gasoline would be required in the market in
2000.  Such a situation presents logistical concerns regarding
the manufacture, supply, and distribution of an additional type
of gasoline (and potentially additional grades), and enforcement
of the entire reformulated gasoline program.   EPA believes that
the petroleum supply and distribution system will already be
strained given the existing reformulated gasoline program.  The
strain could be exacerbated by allowing Phase I gasoline to
endure when Phase II gasoline becomes a requirement for covered
areas in the year 2000.   The argument was also made that
emissions from the increase in tankage and other equipment
necessary for handling additional types and/or grades of gasoline
would reduce the regulation's environmental benefit, although the
extent to which this may occur is unknown.

     Given the potential for supply and distribution problems,
enforcement problems, and the weak statutory authority in the
Act, EPA has decided not to allow Phase I-only opt-ins.

C.   Attainment Area Opt-in
                               473

-------
     A few comments addressed the issue of whether areas in ozone
attainment could opt into the reformulated gasoline program.
EPA's position is that attainment areas may not opt into the
program,  with a possible exception given to areas in an ozone
transport region.

     The opt-in language of §211(k)(6)(A)  of the Act clearly
limits opt-in to areas in a State classified as Marginal,
Moderate, Serious,  or Severe for ozone nonattainment.   One
exception is included in §184(c) of the Act which provides that
upon petition of any State within a transport region established
for ozone, and based on the recommendation of the area's
Interstate Transport Commission, the commission may develop
recommendations for additional control measures to be applied
within all or a part of such transport region if the commission
determines such measures are necessary to bring any area in such
region into attainment.  This provision gives the commission the
opportunity to recommend that an area opt into the reformulated
gasoline program, even those areas in the ozone transport region
which are already in attainment.  Once a commission has made its
recommendation, the usual opt-in application procedures and lead
times apply.

D.   Opt-out

     Once an area has opted into the reformulated gasoline
program,  the issue arises whether it may,  at a later date, decide
to opt out of the program.  This notion was brought up in several
comments, even as a possible alternative to the Phase I-only opt-
in provision discussed above.  While EPA is currently considering
opt-out provisions, section 211(k) does not give EPA the
authority to develop an opt-out procedure.  Thus, EPA is not
including any opt-out provisions in this rulemaking, but may
pursue separate actions in the future.

E.   Opt-In Application

     One commenter asked that EPA develop an opt-in application
procedure.  EPA feels that this procedure is adequately outlined
in §211(k)(6)(A) of the Act, which specifies that the Governor of
the State must apply to the EPA.  EPA has already received a
number of applications in the form of a letter simply requesting
that an area be opted into the reformulated gasoline program.
EPA feels that this process has been an adequate one.   Written
requests for opt-in  may be sent to the Administrator of the
Environmental Protection Agency, West Tower Waterside Mall, 401 M
Street, S.W., Washington B.C.,  20460.
                               474

-------
F.   Opt-In Lead Time

     A few comments requested that extra lead time be given
before allowing areas to opt into the reformulated gasoline
program.  EPA feels that given the existing safeguards in the
Act, such a delay is not warranted.  §211(k)(6)(B) allows EPA to
delay reformulated gasoline in opt-in areas for up to three years
based on insufficient supply.  This is in addition to the
§211(k)(6)(A)  provision that reformulated gasoline not be
required in opt-in areas until January 1, 1995 or one year after
the application is received, whichever is later.   A delay could
be based on a petition from an outside party,  or at EPA's own
initiative.   At the present time, EPA has yet to receive any such
petitions and believes that there will be more than sufficient
supply of reformulated gasoline for a January 1,  1995
implementation date, given the current level of opt-in.

G.   Define Domestic Capacity

     One party's comments contained a request that EPA include
specific definitions for domestic capacity.  EPA is unable to
provide a specific definition of domestic capacity for all areas
at this time.   According to §211(k)(6)(B),  EPA must consult with
the Secretary of Energy before determining whether insufficient
domestic capacity to produce reformulated gasoline exists.  EPA
will define domestic capacity as part of any petition to delay
opt-in.

H.   Opt-in only after Stage II controls and enhanced I/M

     One commenter suggested that opt-in should be allowed only
after an area has implemented Stage II vapor control and an
enhanced inspection and maintenance (enhanced I/M) program.
Given the wide variety of air pollution concerns, EPA believes
that it is of utmost importance to give States the freedom to
implement,  through their State Implementation Plans, whatever
programs they feel will best allow them to achieve attainment.
EPA believes that most opt-in areas will have considered the
benefits of Stage II and enhanced I/M, and some will implement
these programs, but the decision should rest with the States
based on their own needs.  EPA is not comfortable in this
rulemaking preempting State efforts by requiring that Stage II
and enhanced I/M be implemented before allowing opt-in to the
reformulated gasoline program, and hence is not promulgating any
such requirements.
                               475

-------